PD could continue last timeout operator step to avoid useless scheduling

## Enhancement Task
PD's operators will sometimes be timeout when too many snapshot tasks accumulate on TiKV. But by this time, the snapshot task of tikv may have been completed, that is, the learner has applied snapshot. However, pd removes the learner after operator timeout, because the priority of restoring the region to the health status is higher. And then rescheduling.
It is common in replace-offline-peer and replace-down-peer.

Could we continue the previous scheduling step instead of deleting learner.