Background schedule fire loop.

The loop performs bounded FLOW.CLAIM_DUE passes for the internal schedule type. A finite blocking window avoids idle polling while guaranteeing a rescan if another node applied a schedule without delivering a local wakeup. Work remains shard-owned and Raft-guarded; this process only asks for due schedule records and executes the target create/reschedule sequence.

Overdue interval schedules use the schedule's :fire_once catch-up policy. One target is created on recovery, all additional elapsed periods are coalesced in constant time, and the next run is one full interval after the recovery fire. The scheduler never loops once per missed interval.