Worker doesn't hot-reload sync mapping config — requires restart

Context

From #95. `apps/worker/src/build-sync-engine.ts`'s `buildSyncEngine()` is called once at module top-level (`const syncEngine = await buildSyncEngine();` in `apps/worker/src/index.ts`), at process boot. It loads the persisted mapping config once and builds a `SyncEngine` instance that lives for the process lifetime.

If someone edits the mapping config via the `/sync` dashboard while the worker is running, the change has zero effect until the worker process restarts (e.g. next deploy).

Fix options to evaluate

• Periodic re-poll of `SystemConfig` on an interval, rebuilding the `StatusMap`/adapter when the config changes (simplest, adds polling overhead)
• A dashboard-triggered signal (e.g. a Postgres NOTIFY, or a lightweight webhook back to the worker's health-check port) to force a rebuild on save
• Accept the limitation and just surface it clearly in the dashboard UI ("changes take effect after next deploy") instead of building live-reload

No strong opinion yet on which — worth a quick design pass before picking.

Start by reading apps/worker/src/build-sync-engine.ts and apps/worker/src/index.ts, especially the top-level buildSyncEngine() call and how persisted mapping config is loaded. Compare the polling, dashboard-triggered signal, and documented-limitation options, then define which approach should make config changes take effect without a worker restart or clearly surface the limitation in the dashboard.