cron: the heartbeat renewal is awaited inside the select that polls the run, so the run pauses for it #375

Summary

A cron worker drives a running occurrence in a select! over the work, the fence deadline and the heartbeat ticker. The ticker arm awaits renew() — two SQLite UPDATEs — inside the arm body (cron/worker.rs:775-783), so while a renewal is in flight the work future is not polled: the run pauses for the renewal's duration, every heartbeat_interval_secs.

Normally sub-millisecond to a few ms. Under state-database write contention (a busy node, a tmpfs-backed SQLite shared with the scheduler's inserts), it is up to busy_timeout or the fence deadline — a stall inside a turn of a latency-sensitive loop, charged to whatever the run was awaiting (a model_infer deadline counts wall time from the call).

Proposal

Renew concurrently with the work: spawn the renewal (or keep it as a separate pinned future in the select!) so the work keeps being polled, and only the renewal's outcome — lost or fenced — interrupts it.