fix(schedules): persist pending occurrences across retries

[andrei-hasna]

Summary:

• Persist one durable pending occurrence and stable run/turn identity per thread schedule.
• Recover waiting, enqueued, started, and terminal occurrences across retries and restarts.
• Finalize interval, cron, and once cadence plus failure counters exactly once.
• Prevent a reserved scheduled start from overwriting a replacement active turn.

Regression coverage:

• due while busy reuses one occurrence without counting a run or emitting duplicate run state
• duplicate claim and lease fencing
• restart recovery for waiting, enqueued, started, and terminal states
• interval, cron, and once finalization
• cadence and failure counters
• visible scheduled assistant result and persisted turn history

Validation:

• staged diff and credential scans: clean
• Rust formatting check: exit 0
• Blacksmith Testbox package gates: pending
• independent SOL/xhigh adversarial review: pending

Todos plan: 522f0953-22b4-46f6-bca0-50f1bdc2d4eb
Implementation task: 2fe0ce99-23f7-40c4-8769-fc64dfe2ef0b

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}