Process liveness — wired to k8s livenessProbe.

Returns 200 while the Python process is alive. Has no dependency check by design: restarting a pod can't fix a Supabase outage, and the thrashing only deepens an incident. See ARD-728 and app/health.py for context.

Declared async deliberately (ARD-2136). A sync def endpoint is dispatched to Starlette's threadpool, which is the same pool check_supabase_ready occupies via asyncio.to_thread on every 5s readiness probe. Liveness then queues behind dependency work it is documented not to depend on, times out, and the pod is killed anyway — exactly the thrashing ARD-728 set out to prevent. Answering on the event loop removes that coupling.