[FLINK-29840] Fix bug that old record may overwrite new record in Table Store when snapshot committing is slow

[tsreaper]

(Cherry-picked from #354)

Consider the following scenario when snapshot committing is slow:

• A writer produces some records at checkpoint T.
• It produces no record at checkpoint T+1 and is closed.
• It produces some records at checkpoint T+2. It will be reopened and read the latest sequence number from disk. However snapshot at checkpoint T may not be committed so the sequence number it reads might be too small.

In this scenario, records from checkpoint T may overwrite records from checkpoint T+2 because they have larger sequence numbers.

This PR fixes this bug by comparing last modified commit identifier and the latest committed identifier before closing a writer. If by comparing we found that the last modification is already committed, we can safely close the writer.