Make stale scheduled-claim restart test deterministic

[owainlewis]

Problem

The Ubuntu PR run for #178 intermittently failed in jobs::tests::live_scheduled_claim_survives_restart_and_stale_claim_becomes_deliverable at src/jobs.rs:5391. The test expected the recovered row status to be failed, but it was still running. The same commit passed locally, on macOS CI, and on an unchanged Ubuntu retry.

Run: https://github.com/owainlewis/push/actions/runs/30565403853

Outcome

Make the lease-expiry and restart scenario deterministic without weakening the production recovery assertion.

Acceptance criteria

• The test controls or explicitly advances the relevant lease clock instead of depending on runner timing.
• It still proves that a live scheduled claim survives restart and that a stale claim becomes deliverable.
• Focused repeated runs pass on Linux and macOS.
• The full locked test suite passes.

Checks

• Run the focused test repeatedly under constrained CI-like timing.
• Run cargo test --locked.
• Run formatting and clippy.

[tomatotomata]

I’m taking this as a test-only determinism slice. I’ll trace the existing lease clock, restart harness, and stale-claim test first, then replace runner-timing assumptions with an explicit clock/control point while keeping the recovery assertion intact. I’ll run the focused test repeatedly, cargo test --locked, formatting, and clippy, and I’ll keep production behavior unchanged. Let me know if the project prefers a test-clock injection or an existing scheduler test hook once I map the current helpers.