E2E: crash recovery — succeeded jobs skipped on restart

**Epic:** #1082

During a `frozen`-phase job execution, kill the daemon process. Restart it. Assert:
- Already-completed jobs are skipped (idempotency label query returns Succeeded)
- Remaining jobs execute normally
- Upgrade completes

## Acceptance Criteria
- [ ] Skipped job is not re-launched (no duplicate K8s Job created)
- [ ] Upgrade reaches Succeeded state after restart
- [ ] Test covers both daemon-side (frozen) and UC-side (prepare) crash scenarios

Start with the E2E harness and the daemon and UC execution paths named in the issue; identify how frozen-phase and prepare-phase crashes are injected and restarted. Run the relevant upgrade scenario, then verify the idempotency label query skips succeeded work, creates no duplicate Kubernetes Job, executes remaining jobs, and reaches Succeeded.

Written by the indexing model from the issue text.