[platinumhamburg]

Purpose

Closes #4952.

After successful Undo Recovery, an idle bucket keeps its original checkpoint baseline even though recovery has appended compensating writes. Subsequent checkpoints can therefore continue to depend on old changelog history and fail to restore after that history expires.

Brief change log

• Return acknowledged log end offsets from Undo writes and retain the maximum per bucket after all writes complete.
• Use the scanned recovery boundary when no compensating writes are required.
• Persist these recovered baselines in subsequent checkpoints without requiring new sink records.

This does not extend the recoverability of older checkpoints beyond the available changelog history.

Tests

• Reproduced the idle-bucket regression on main: the saved baseline remained 1 after Undo advanced the log end to 5.
• Local clean build: 57 tests passed across UndoRecoveryOperatorStateTest, UndoRecoveryExecutorTest, UndoComputerTest, RecoveryOffsetManagerTest, UndoRecoveryOperatorFactoryTest, and UndoRecoveryManagerITCase.
• Coverage includes a checkpoint and second restore without new sink writes, out-of-order write acknowledgements, and recovery ranges containing only empty batches.

API and Format

No changes to connector options or checkpoint serialization.

Documentation

No documentation changes required.

[wuchong]

LGTM.

Please resolve the merge conflicts.