gc_worker: reduce GC scan locks when meeting region cache miss (#18385)

[ti-srebot]

cherry-pick #18385 to release-4.0

---

What problem does this PR solve?

Issue Number: close #xxx

Problem Summary:

currently, gc worker will rescan locks when meet region cache miss error which makes slow scan job become more slow.

What is changed and how it works?

What's Changed, How it Works:

when region cache miss error(current is a fake epochNotMatch), it just indicates "maybe region has changed and we need to reload region from PD", we can continue resolve scan lock result if all lock still in that region.

this can help situations like:

• lock and region hasn't changed but region cache expired
• lock's region's key range didn't change but maybe add new peer(id is same but confVer or ver changed)
• all lock has changed from region A to region B

but it didn't help the situation that:

• lock's region has been split into two region

maybe we can improve it later(to avoid complex modification)

Related changes

• Need to cherry-pick to the release branch

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)
• No code

Side effects

• n/a

Release note

• Reduce GC scan locks when meeting region cache miss

---

This change is 

[ti-srebot]

/run-all-tests

[jackysp]

LGTM

[jackysp]

/merge

[ti-srebot]

Your auto merge job has been accepted, waiting for:

• 18529

[jackysp]

/run-unit-test