What problem does this PR solve?

• The start-ts of a table could regress if the table is added to a capture whose current checkpoint-ts is less than the start-ts of said table, and before the checkpoint-ts catches up, the capture exits abnormally. This regression, in rare cases, could cause the start-ts to regress so far that it is earlier than the global checkpoint-ts. When this happens, SchemaStorage related errors occur. checkpointTs invariant violation after CREATE TABLE #1258 meet ErrSchemaSnapshotNotFound error in TiCDC 4.0.9 #1235

What is changed and how it works?

• Fixed several bugs in Owner so that never will a local checkpoint be less than the global checkpoint.
• Cleaned up checkpoint ts semantics with respect to DDL operations, so that a DDL is finished if and only if global checkpoint >= FinishTs.
• Fixed a bug in processor introduced by *: fix the output in changefeed out of order #1247 which causes local checkpoint regression.
• Added Etcd state checker to integration test.

Check List

Tests

• Unit test
• Integration test

Related changes

• Need to cherry-pick to the release branch

Release note

• Fix bugs in the owner that could lead to abnormal CDC server exits during table migrations.