fix(scheduler): address !7 REQUEST_CHANGES

- seed: check completed_key before no_source so same-day last_completed
  is already_completed (not 10m retry)
- scheduler: resolve TaskControl ok/error before logging; ok=False logs
  task_fail, not task_ok
- candidate_cleanup: catch failures, record task, retry ~5m like discover

reaction_board/cli.py

@@ -730,10 +730,20 @@ def cmd_run(args, logger: JsonLogger, *, serve_web: bool = False) -> int:
                 return _quiet_window_defer(
                     repo, logger, "candidate_cleanup", started_at=started,
                 )
-            removed = repo.delete_expired_candidates(
-                ttl_days=cfg.schedule.candidate_ttl_days
-            )
-            expired = repo.expire_fav_actions()
+            try:
+                removed = repo.delete_expired_candidates(
+                    ttl_days=cfg.schedule.candidate_ttl_days
+                )
+                expired = repo.expire_fav_actions()
+            except Exception as exc:  # noqa: BLE001  记录后约 5 分钟补跑
+                finished = int(time.time())
+                _record_task(
+                    repo, "candidate_cleanup", ok=False, started_at=started,
+                    finished_at=finished, error=str(exc),
+                )
+                return TaskControl(
+                    ok=False, next_run=_retry_soon(float(finished)), error=str(exc),
+                )
             if removed or expired:
                 repo.bump_cache_version("videos")
             finished = int(time.time())