Complete hunt-job lifecycle with leases, retries, cancellation, and dead letters

[haasonsaas]

Context

The hunt_job table currently records enqueue and claim metadata and is described as supporting multi-worker claim in a later step. Retro-hunting needs operational controls when a worker crashes, a rule is invalid, or a tenant cancels a long-running job.

Proposal

Implement an explicit hunt-job state machine with lease expiry, heartbeat/renewal, retry budget, backoff, cancellation, terminal failure, dead-letter reason, and progress counters.

Acceptance criteria

• A crashed worker's job becomes claimable after a bounded lease timeout.
• Retries and terminal failures are idempotent and queryable.
• Cancellation prevents new work and reports already-processed results.
• The API/CLI exposes progress, last error, claimed worker, and retry state.

References

• hunt queue schema
• hunt core module