Latency accelerator for delayed jobs. On a short interval it finds queues with work due to run now (a job's scheduled_at has elapsed, including a failed job rescheduled after its backoff) and issues a pg_notify('shigoto_jobs_insert', Queue) so the existing notifier wakes the matching queue immediately instead of waiting up to poll_interval.

Uses pg_try_advisory_xact_lock so only one node stages in a multi-node deployment; the notification still reaches every node.

Started only when the notifier is configured. The per-queue polling fallback remains the correctness backstop; the stager only reduces latency, and fails soft on any database error so a transient fault never crashes the supervision tree.