[Feature] Make abort a reliable recovery path for wedged terminating jobs #4343

Summary

For a job wedged in TERMINATING, abort: true may update the termination reason without causing the job to be reprocessed or clearing an expired worker lock. Operators then have no run-level recovery path and may need a separate fleet-instance deletion operation.

Expected behavior

Provide a documented, idempotent force-recovery operation that safely requeues or finalizes the selected job, handles stale locks, records an event/result, and cannot terminate an instance still serving unrelated active jobs.

Acceptance criteria

• The caller can target a specific run/job and inspect whether force recovery took effect.
• Repeated requests are safe.
• Active shared-instance workloads are protected.
• Tests cover a stale terminating job, an active job, and a failed recovery attempt.

This is distinct from the older abort/container report #893; the concern here is a server-side job-processing wedge.