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/scheduler.py

@@ -87,11 +87,9 @@ class Scheduler:
                 raw = task.fn()
             except Exception as exc:  # noqa: BLE001  单任务失败不终止调度
                 error = str(exc)
-                self._log("error", "scheduler.task_error", task=task.name, error=error)
             else:
                 if isinstance(raw, TaskControl):
                     control = raw
-                self._log("info", "scheduler.task_ok", task=task.name)
             finished = self._clock()
             if control is not None and control.next_run is not None:
                 task.next_run = float(control.next_run)
@@ -104,6 +102,23 @@ class Scheduler:
                 ok = bool(control.ok) and error is None
                 if control.error and error is None:
                     error = control.error
+            # Skip/fail ≠ success：先定 ok/error，再记 scheduler 日志。
+            if ok:
+                self._log("info", "scheduler.task_ok", task=task.name)
+            elif control is not None:
+                self._log(
+                    "warning",
+                    "scheduler.task_fail",
+                    task=task.name,
+                    error=error,
+                )
+            else:
+                self._log(
+                    "error",
+                    "scheduler.task_error",
+                    task=task.name,
+                    error=error,
+                )
             result = TaskResult(
                 name=task.name,
                 started_at=started,