fix(live): translate IMEI↔UUID in snapshot + device-event-map joins

positions.device_id stores the IMEI (text); entry_devices.device_id is a
uuid FK to devices.id. snapshot.ts and device-event-map.ts joined the two
columns directly, causing:

  - snapshot.ts: Postgres rejected with `operator does not exist: uuid =
    text` (42883). The registry caught the error and returned an empty
    snapshot, masking the failure.
  - device-event-map.ts: cache keyed on entry_devices.device_id (uuid),
    but broadcast.ts:141 looks up by position.device_id (imei). Cache
    missed every record → no live frames fanned out, silently.

Both queries now hop through the devices table (devices.imei =
positions.device_id, devices.id = entry_devices.device_id). The
device-event-map cache aliases d.imei AS device_id so cache keys stay
IMEI strings — broadcast.ts is unchanged.

The integration-test fixture schema previously had entry_devices.device_id
as text (IMEI) — a deliberate simplification that hid the production type
mismatch. Now matches production: adds a devices table and changes the FK
to uuid. seedDatabase inserts devices first.

178/178 unit tests pass. Integration test exercises the corrected join
shape.

test/fixtures/test-schema.sql

@@ -1,39 +1,45 @@
 -- test/fixtures/test-schema.sql
 --
 -- Minimum subset of the production schema required by live.integration.test.ts.
--- This is intentionally a simplified version — NOT the full Directus-managed schema.
+-- Intentionally a simplified subset of the Directus-managed schema — keeps only
+-- the columns the Processor's live-broadcast queries actually read.
 --
--- Maintenance note: keep in sync with the real schema when column types change on
--- these tables. Specifically: entries.event_id, entry_devices.device_id (Phase 1
--- uses IMEI text; Phase 2 introduces UUID-based devices table).
---
--- Phase 1 deviation: entry_devices.device_id is TEXT (IMEI) here, matching
--- positions.device_id. The real Directus schema uses a UUID FK to devices.id.
--- The integration test uses the real queries from device-event-map.ts and
--- snapshot.ts, so this simplified schema must satisfy those joins.
+-- Maintenance note: keep in sync with the real schema when join shapes change.
+-- The integration test runs the real queries from device-event-map.ts and
+-- snapshot.ts unmodified, so the column types here must match production.
 
--- events — the container for entries
--- The Processor reads events.id (used in snapshot WHERE e.event_id = $1).
+-- events — the container for entries.
+-- Processor reads events.id (snapshot WHERE e.event_id = $1).
 CREATE TABLE IF NOT EXISTS events (
   id   uuid PRIMARY KEY DEFAULT gen_random_uuid()
   -- Real schema also has: organization_id FK, name, slug, discipline, starts_at, ends_at.
-  -- Only columns the Processor queries are included here.
 );
 
--- entries — race entries belonging to an event
--- The Processor reads entries.id and entries.event_id.
+-- entries — race entries belonging to an event.
+-- Processor reads entries.id and entries.event_id.
 CREATE TABLE IF NOT EXISTS entries (
   id       uuid PRIMARY KEY DEFAULT gen_random_uuid(),
   event_id uuid NOT NULL REFERENCES events (id) ON DELETE CASCADE
-  -- Real schema also has: vehicle_id, class_id, number, etc.
+  -- Real schema also has: vehicle_id, class_id, race_number, status, etc.
 );
 
--- entry_devices — maps a device (IMEI) to an entry.
--- Phase 1: device_id is IMEI text, matching positions.device_id.
--- Real schema: device_id is UUID FK to devices.id, joined via devices.imei.
--- This simplified form is intentional for the integration test fixture.
+-- devices — durable hardware catalog.
+-- Processor reads devices.id and devices.imei (joined to positions.device_id).
+-- positions.device_id stores the IMEI text; entry_devices.device_id stores the
+-- devices.id uuid. This table is the bridge that lets snapshot.ts and
+-- device-event-map.ts translate between the two without `uuid = text` errors.
+CREATE TABLE IF NOT EXISTS devices (
+  id   uuid PRIMARY KEY DEFAULT gen_random_uuid(),
+  imei text NOT NULL UNIQUE
+  -- Real schema also has: model, serial_number, notes, date_created, date_updated.
+);
+
+-- entry_devices — maps a device (uuid FK) to an entry.
+-- Real schema (and this fixture): device_id is uuid FK to devices.id.
+-- Live-broadcast joins translate to/from positions.device_id (IMEI text)
+-- via the devices table.
 CREATE TABLE IF NOT EXISTS entry_devices (
   id         uuid PRIMARY KEY DEFAULT gen_random_uuid(),
   entry_id   uuid NOT NULL REFERENCES entries (id) ON DELETE CASCADE,
-  device_id  text NOT NULL  -- IMEI in Phase 1
+  device_id  uuid NOT NULL REFERENCES devices (id) ON DELETE CASCADE
 );

test/live.integration.test.ts

@@ -63,9 +63,11 @@ const BROADCAST_GROUP_PREFIX = 'live-broadcast';
 const EVENT_ID = 'ee000000-0000-0000-0000-000000000001';
 const OTHER_EVENT_ID = 'ee000000-0000-0000-0000-000000000002';
 const ENTRY_ID = 'aa000000-0000-0000-0000-000000000001';
-const DEVICE_1 = '111111111111111'; // IMEI
-const DEVICE_2 = '222222222222222'; // IMEI
-const DEVICE_ORPHAN = '999999999999999'; // not in entry_devices
+const DEVICE_1_ID = 'dd000000-0000-0000-0000-000000000001';
+const DEVICE_2_ID = 'dd000000-0000-0000-0000-000000000002';
+const DEVICE_1 = '111111111111111'; // IMEI for DEVICE_1_ID
+const DEVICE_2 = '222222222222222'; // IMEI for DEVICE_2_ID
+const DEVICE_ORPHAN = '999999999999999'; // not registered to any entry
 
 const USER_A: FakeUser = {
   id: 'user-aaaa-0000-0000-0000-000000000001',
@@ -220,12 +222,20 @@ async function seedDatabase(pool: pg.Pool): Promise<void> {
     [ENTRY_ID, EVENT_ID],
   );
 
-  // entry_devices — Phase 1 uses IMEI as device_id
+  // devices — durable catalog. positions.device_id stores the IMEI text;
+  // entry_devices.device_id is a uuid FK to devices.id. The live-broadcast
+  // joins translate via devices.imei.
+  await pool.query(
+    `INSERT INTO devices (id, imei) VALUES ($1, $2), ($3, $4)`,
+    [DEVICE_1_ID, DEVICE_1, DEVICE_2_ID, DEVICE_2],
+  );
+
+  // entry_devices — uuid FK to devices.id (matches production schema).
   await pool.query(
     `INSERT INTO entry_devices (id, entry_id, device_id) VALUES
       (gen_random_uuid(), $1, $2),
       (gen_random_uuid(), $1, $3)`,
-    [ENTRY_ID, DEVICE_1, DEVICE_2],
+    [ENTRY_ID, DEVICE_1_ID, DEVICE_2_ID],
   );
 
   // positions for DEVICE_1 (two: one non-faulty, one faulty)