feat(scheduler): heartbeat-aware health check for scheduled sync - #9

Why

The HTTP-based image HEALTHCHECK is only suitable for the web process. PR #8 disabled it for scheduler as a short-term correction. This follow-up adds a scheduler-specific, non-HTTP heartbeat/deadline check.

Implementation

• Atomically record non-sensitive scheduler state in data/scheduler_heartbeat.json (0600): PID, start/heartbeat timestamps, sleep vs running, next scheduled run, previous completed run start/end and return code.
• Wake from normal daily sleep every 30 seconds to pulse; pulse during long interval-sync subprocess waits.
• Check process existence, heartbeat freshness (180 sec), overdue scheduled run (15-minute grace), and max running duration (4 hours).
• Preserve last completed result while next run is in progress; failed remote sync is reported, not conflated with scheduler liveness.
• Do not overwrite persistent heartbeat for --once jobs sharing the same mounted data directory.
• Replace PR fix(compose): disable inherited healthcheck for scheduler #8's scheduler.healthcheck.disable: true with a Python-based Compose healthcheck. Keep the web image's HTTP healthcheck unchanged.
• Add unit tests and operational README instructions.

Deployment and caution

This is a separate follow-up to merged PR #8, not a hotfix to the currently running OCI container. Rebuild the image and recreate services only after CI and review. Docker's unhealthy status does not automatically restart containers; configure monitoring separately.

Validation

GitHub CI: Python tests, frontend build, Docker smoke, history secret scan.