Summary

Two small but impactful fixes for cron ticker reliability and observability.

1. cron/scheduler.py — Narrow file lock scope + fire-and-forget execution

Problem: The file lock in tick() covered the entire cycle — scheduling AND job execution. A long-running job would block subsequent ticks, potentially delaying or missing future job deadlines in high-frequency schedules.

Fix:

• Narrow the file lock to only the scheduling phase (get_due_jobs + advance_next_run), then release it before any job execution begins.
• Use fire-and-forget pattern (ThreadPoolExecutor.submit + shutdown(wait=False)) for parallel jobs, so tick() returns immediately instead of blocking on job completion.

This means subsequent ticks are never blocked by running jobs — the lock only serializes the scheduling bookkeeping.

2. gateway/run.py — Heartbeat + robust error logging

• Heartbeat file: Write a timestamp to ~/.hermes/cron/.ticker_heartbeat every 5 ticks (~5 minutes). Enables external watchdogs to detect a dead ticker process.
• Heartbeat log: Log an INFO-level "Cron ticker heartbeat" message at the same interval, so agent.log always shows the ticker is alive.
• Exception → BaseException: All except Exception blocks changed to except BaseException with exc_info=True, so SystemExit/KeyboardInterrupt/other critical signals are logged with full traceback instead of being silently swallowed.

Testing

• tick() runs scheduling under lock, releases before execution
• Fire-and-forget parallel execution does not block subsequent ticks
• Heartbeat file written every 5 ticks
• Heartbeat log emitted at correct interval
• All existing cron functionality preserved