ddl_queue + mutation is no longer reliable way to confirm migration success

Occasionally, the system goes into a state where the 100(x) migration status is in progress, but the 100(x+1) succeeded. This doesn't make sense, because we only run 100(x+1) if there is no pending mutation or entry in the queue. Either because the pod is getting killed or something around that, or the way we conclude the migration status is no longer reliable.