What problem does this PR solve?

Issue Number: close #1787

Problem Summary:

1. Sometime the panel Region write Duration (write blocks) appear in weird places. It is because the panel share a same "id=77" with the panel Handling Request Number
2. After restored DTFiles (single file mode), we may get inconsistent rows from a DTFile even if we read with a same key range. Because the min-max indexes can not be loaded successfully.

What is changed and how it works?

1. Fix the id of panel and re-export the json file for TiFlash-Summary
2. Do DMFile::initializeSubFileStatIfNeeded inside DMFilePackFilter::init to ensure we can load min-max indexes if they exist.
3. Do DMFile::initializeSubFileStatIfNeeded inside DMFile::readMeta.
   @lidezhu split the functions from readMeta to reduce the number of bytes read from the disk while initiating. But it brings another problem: We need to acquire a lock for checking whether the file is single-mode and sub_file_stats is empty every time we create a DMFileFilter. No matter the file is single-mode or directory-mode.
   Those data are small and we can optimize the initlizating by lazy init DeltaMergeStore.

Related changes

• PR to update pingcap/docs/pingcap/docs-cn:
• Need to cherry-pick to the release branch: 5.0

Check List

Tests

• Unit test

Side effects

• N/A

Release note

• Fix the bug that the Grafana panel Region write Duration (write blocks) may be shown in wrong place