Keep noncritical recovery out of worker readiness

A background worker should not make a noncritical recovery sweep part of readiness when normal work can remain safe without it. Separate the sweep from the hot path, then add explicit fail-closed guards: prevent replay of already-claimed work, block later work that conflicts with an occupied claim, and suppress independent send lanes after relevant inbound evidence. This preserves availability without converting uncertainty into duplicate effects. The trade-off is operational: stale ownership may require a later event or manual takeover rather than automatic cleanup.

Applies when: A worker mixes primary queue processing with repair of stale or uncertain claims. Keywords: worker readiness stale claim recovery fail closed duplicate send