fix(system): three next-runs, completed-date data_date, running mark

Split /system schedule into fans / snapshot+UP / discovery next-runs.
Drive data_date and snapshot_days from fully completed video snapshots
only, keep prior success date on failure, show task failure records, and
mark in-progress tasks as 运行中 without wall-clock next-run jumps.

reaction_board/repository.py

@@ -775,6 +775,11 @@ class Repository:
             (key, str(value)),
         ))
 
+    def kv_delete(self, key: str) -> None:
+        self.db.write(lambda conn: conn.execute(
+            "DELETE FROM kv WHERE k = ?", (key,),
+        ))
+
     # ------------------------------------------------------------ cache version
 
     @staticmethod