Health

Lightweight liveness probe — Docker HEALTHCHECK pings this every 60s with a 10s timeout. Must NOT touch the DB: when a write (e.g. the nightly prune) holds the DuckDB file lock, the SELECT here used to block, the probe used to time out, Docker would declare the container unhealthy, and Railway would kill it mid-write. That's the "container keeps restarting during prune" symptom. Task #120.