When the status of learner is abnormal, scheduling cannot be performed.

## Bug Report

### What did you do?
There is a cluster, and a placement rule with learner role was previously configured. But the corresponding label was not added to TIKV due to insufficient machine resources. So there is no learner peer. Now a large amount of data is written to a table in the cluster.

### What did you expect to see?
As the data increases, the region splits and balance region to all TikV.

### What did you see instead?
It is found that the storage used size of three TIkV nodes continues to increase, and the `balance-region` scheduler is not working normally.

### What version of PD are you using (`pd-server -V`)?
4.0.14