Periodically returns executing rows with an expired lease to runnable, bumping attempt. This is the at-least-once safety floor for worker crashes; handle/2 is intentionally not involved.

The same tick also fires await timeouts: parked rows whose await_deadline passed go back to runnable (a wake, not a failure — attempt untouched), so timeout resolution is bounded by the sweep interval.

Configured per node via the engine's :reaper option ([interval: 30_000], or false to not run a reaper on this node — at least one node in the cluster must, or the leases of crashed workers are never reclaimed).