Fix status file race condition (read-modify-write is not concurrency safe)
Problem: update_status_file() is atomic only at os.replace(), but the read-modify-write sequence is not concurrency safe. Multiple writers can overwrite each other and lose fields. Tasks: Choose a strategy: enforce single-writer per job OR add file locking OR switch to append-only event log. Implement and add stress test. Acceptance: Concurrent updates do not lose fields. Status JSON remains valid under stress. Priority: High.