restore: apply adjust max-pending-peer-count when stop pd schedulers

[glorv]

What problem does this PR solve?

Apply the change of max-pending-peer-count for pd scheduler in https://github.com/pingcap/br/pull/644 to avoiding scatter region failure.

What is changed and how it works?

Check List

Tests

• Unit test
• Integration test
• Manual test
  • Manual run 14k warehouse tpcc benchmark with 6tikv. After this patch, region&leader&store distribution are always even in the whole procedure.
• No code

Side effects

Related changes

[glorv]

Waiting for https://github.com/pingcap/br/pull/644

[lance6716]

LGTM

[glorv]

/merge

[ti-srebot]

/run-all-tests