This is an manual cherry-pick of #8701 and #8698

What problem does this PR solve?

Issue Number: close #8695

Problem Summary: As the issue describe: #8695 (comment)

What is changed and how it works?

Introduce a GlobalPageIdAllocator to avoid a page_id from being reused in one physical_table_id scope. Specifically,

• Introduced a GlobalPageIdAllocator to allocate page_id for segment_id, delta-layer and stable-layer for all IStorage instances instead of each StoragePool allocate the page_id for its own IStorage instance independently
• Each IStorage instance restored from disk, it will first try raise the lower bound of the GlobalPageIdAllocator. After that, a table won't be able to reuse the allocated id
• If an IStorage instance being physically removed (cause by setting tiflash replica to 0) and re-create again, it will still allocated the page_id from GlobalPageIdAllocator, so no page_id will be reused too.

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)

Load tpc-ds 50
Set tiflash replica 1 for database tpcds50 and wait for all tables progress == 1
Set tiflash replica 0 and wait for gc
Set tiflash replica 1 and wait for all tables progress == 1
Run queries on tpcds 50 through tiflash

• No code

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Documentation

• Affects user behaviors
• Contains syntax changes
• Contains variable changes
• Contains experimental features
• Changes MySQL compatibility

Release note

Fix the issue that the tiflash replica data may be corrupted after setting the tiflash replica to 0 and add it back later