feat(sites): one install, many networks (v0.24.0-beta)

Closes #130. Multi-site, not multi-tenancy: a site is a named collection of
devices whose clients, events, topology and dashboards are viewed separately.
Sites separate what you see, not who you are — the admin/manage/read-only
roles are unchanged and every user sees every site. Per-site permissions
would have multiplied the work for a need nobody stated.

Two measurements decided the design.

Every device-derived table hangs off devices(id) — 25 of 50 tables, 35
foreign keys — so a single site_id column on devices scopes clients, events,
topology, backups and the rest transitively. No other table needed one.

The frontend has 162 useQuery call sites across 57 files. Threading a site
through each would be a large change that leaves a permanent trap: every
future query another chance to forget. The site is injected in the existing
axios interceptor instead, beside the auth token, so a new query is scoped by
default. Only 23 collection routes needed filtering; the 131 per-device
routes are already scoped by their device.

Existing installs are untouched. Migration creates one "Default Site", adopts
every device into it, and a single-site install never meets the concept. An
absent X-Site-Id header means unscoped, so existing API tokens and scripts
behave exactly as before.

Polling stays fleet-wide. Sites are a view over data already gathered, not a
change to how it is gathered — a device does not stop being monitored because
you are looking elsewhere. Rogue-AP detection scopes the scans but keeps the
"these are ours" BSSID set fleet-wide, or your own AP in another site reads
as an evil twin. Bulk actions do obey the selector: an operation that
silently spanned two customers is the blast radius the wave guards exist to
contain.

Four rounds of defects came out of testing the UI rather than the API, and
the first is the one worth recording. Switching sites called
queryClient.clear(), which empties the cache and reviews as correct. It
isn't: clear() destroys each query but a *mounted* observer keeps its own
reference and last result, so every rendered page went on showing the
previous site's data. The QueryClient is now keyed on the site, which
remounts the tree against an empty cache. Three tests pin the old behaviour
down so the shortcut cannot return.

The others: an unscoped fallback path behind the Connected Clients chart;
Config Health, TX retries and the activity feed reading fleet-wide; and
site_id added to the device list query but none of the other four that build
a device payload, so the Site field could not read its own value and appeared
to reject every change while the writes were landing. That last one is now
structural — DEVICE_BASE_COLUMNS is shared by all five, with a test that
fails when a device payload is hand-rolled, and which found two further
instances while being written.

Frontend tests were never run by CI despite vitest being configured and six
suites existing; they now run in both the workflow and ci-preflight.

Verified against a live four-device fleet: the migration left every figure
identical, an empty site returns zero from every read endpoint, moving a
device carries its history with it, and deleting a site with devices is
refused.

Minor bump rather than patch: sites change the shape of the product rather
than adding to it.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/README.md

@@ -6,6 +6,7 @@ whether to try it, the [project README](https://github.com/2GT-Media-Group-LLC/m
 | Document | Covers |
 |---|---|
 | [Configuration](configuration.md) | Environment variables, secret management and key rotation |
+| [Sites](sites.md) | Grouping devices into sites, switching between them, and what each site scopes |
 | [Change Guard and Config Health](change-guard.md) | How the safety system works, its settings, and what to expect when it fires |
 | [Alerting](alerting.md) | Alert rules, delivery channels, webhooks, scheduled reports, maintenance windows |
 | [Single sign-on (OIDC)](sso-oidc.md) | Identity provider setup, group-to-role mapping, break-glass behaviour |

docs/api.md

@@ -43,6 +43,25 @@ or 2FA changes.
 | `GET /api/devices/:id/config-health` | Latest standing-audit findings |
 | `GET /api/topology` | Graph of devices, links, external nodes, and distrusted identifiers |
 | `GET /api/operations/insights` | The dashboard's "things to handle" feed |
+| `GET /api/sites` | Sites with device counts (see [Sites](sites.md)) |
+
+### Scoping a request to one site
+
+Collection endpoints accept an `X-Site-Id` header (or `?site_id=`) that limits the
+response to devices in that site:
+
+```bash
+curl -H "Authorization: Bearer $TOKEN" -H "X-Site-Id: 2" \
+     https://mtm.example.com/api/devices
+```
+
+**Omitting it means unscoped**, returning the whole fleet exactly as before sites
+existed — existing scripts and tokens need no change. Per-device endpoints
+(`/api/devices/:id/...`) are already scoped by the device itself and ignore the header.
+
+One thing to know if you create devices via the API: a device joins the site the
+request was scoped to, falling back to the default site. A device belonging to no site
+is invisible whenever a site is selected, so this is set rather than left null.
 
 `preflight` is the useful one for automation: it returns the same verdict the UI shows,
 so a pipeline can refuse its own change before touching the device. See

docs/sites.md

@@ -0,0 +1,127 @@
+# Sites
+
+[← Documentation index](README.md)
+
+A **site** is a named collection of devices whose data — clients, events, topology,
+backups, statistics — is viewed separately from every other site's.
+
+This is multi-*site*, not multi-tenancy. Sites separate what you **see**; they do not
+separate who you **are**. There is deliberately no per-site access control: the existing
+admin / manage / read-only roles are unchanged, and every user can see every site.
+
+## If you run one network, ignore this page
+
+On upgrade, a single site named **Default Site** is created and every existing device is
+assigned to it. Nothing moves, no figure changes, and the site selector does not appear.
+You will never meet the concept unless you create a second site.
+
+## Creating and switching
+
+The selector sits under the logo in the top-left, below the brand block. It lists your
+sites and offers **New site**, **rename** (the pencil, on hover) and **All sites**.
+
+Switching site reloads the data you are looking at. This is intentional and not a
+performance bug: every cached figure belonged to the site you just left, and showing it
+under another site's name would be worse than a brief reload.
+
+## Moving devices
+
+A device belongs to exactly one site. Move it from **Device → Physical Details → Site**;
+the picker only appears when more than one site exists.
+
+Moving a device takes its history with it — clients, events, backups, interface
+statistics. Nothing is copied or lost, because none of that data carries a site of its
+own: it hangs off the device, and the device carries the site.
+
+Devices added through the UI or API join whichever site you were viewing, falling back
+to the default site. A device belonging to no site would be invisible the moment any
+site is selected, so the field is always set rather than left empty.
+
+## The all-sites view
+
+**All sites** in the selector opens a table of every site with its address and device
+count, plus a world map with a pin per geocoded site. Selecting a row switches to that
+site.
+
+Choosing **All sites** also makes every other page fleet-wide, which is the honest reading
+of "all sites" — useful for a fleet-wide search, less so for a dashboard that averages
+several unrelated networks.
+
+Pins require an address, which is geocoded via Nominatim when you save it. If
+**Settings → Maps enabled** is off, no geocoding or tile request is made and the map is
+hidden; addresses are still stored and shown as text.
+
+## What sites scope, and what they do not
+
+Scoped to the selected site:
+
+- Devices, clients, events, topology, search
+- Dashboard and Operations insights, metrics summary
+- Wireless (APs, CAPsMAN, RF channels and signals)
+- Backups, Firmware, Bulk commands
+
+Also scoped: the Operations feed (Things to Handle, Recent Activity, capacity), Config
+Health findings, anomaly detection, and the Connected Clients chart.
+
+Deliberately **not** scoped:
+
+- **Polling.** Collection always runs across the whole fleet. Sites are a view over data
+  already gathered, not a change to how it is gathered — a device does not stop being
+  monitored because you are looking at another site.
+- **The audit log**, in Recent Activity. Audit rows record who changed the install; they
+  carry no device and so belong to no site. Repeating "admin changed a setting" into every
+  site would misattribute it, so the audit feed appears only in the all-sites view.
+- **NetFlow collector health**, which is one service for the install.
+- **Rogue AP detection.** The scans are site-scoped, but the "these access points are
+  ours" set stays fleet-wide. An AP of yours in another site is still yours, and scoping
+  that set would report your own hardware as a rogue.
+- **Install-wide configuration**: users, alert rules, webhooks, config templates,
+  credential presets, tags, maintenance windows and settings.
+
+Bulk actions — Backup all, Sync all, firmware check-all, bulk commands — obey the
+selector. A bulk operation that silently spanned two customers is precisely the blast
+radius the wave and halt-on-failure guards exist to contain.
+
+## Deleting a site
+
+Deleting a site never deletes the devices in it. If devices remain, the deletion is
+refused and you are asked to move them first, so their destination is a decision rather
+than a side effect.
+
+## API
+
+The active site travels as an `X-Site-Id` header, or a `?site_id=` query parameter.
+
+```bash
+curl -H "Authorization: Bearer $TOKEN" -H "X-Site-Id: 2" https://mtm.example.com/api/devices
+```
+
+**Omitting the header means unscoped**, returning the whole fleet exactly as the API did
+before sites existed. Existing scripts and API tokens keep working untouched.
+
+| Method | Path | Purpose |
+|---|---|---|
+| `GET` | `/api/sites` | List sites with device counts |
+| `POST` | `/api/sites` | Create a site |
+| `PUT` | `/api/sites/:id` | Rename, set address or notes |
+| `DELETE` | `/api/sites/:id` | Delete (refused while devices remain) |
+| `POST` | `/api/sites/:id/devices` | Move devices into the site |
+
+## The Connected Clients chart
+
+Client totals are deduplicated — a client seen by two access points is one client — so the
+chart cannot be a sum of per-device series. A per-site deduplicated series is recorded
+alongside the fleet-wide one, and it starts accumulating from the upgrade.
+
+That would have cost every existing install its chart history, so there is one shortcut,
+and it is an identity rather than an approximation: **when a site contains every device,
+its total is the fleet total**, and the chart reads the fleet-wide series with all of its
+history. Single-site installs therefore lose nothing. Once you genuinely split the fleet,
+each site's chart builds its own history from that point.
+
+## A note on client counts
+
+A client is counted in a site when a device in that site has seen it. On genuinely
+separate networks the sites do not overlap. If two sites share one broadcast domain —
+common in a lab — the same MAC is legitimately seen by hardware in both, and appears in
+both counts.