Remove table and clean up redundant state info when scanning newly added table

[qidian99]

As mentioned in Issue#913, when a task is stopped and its configuration changed, there might be redundant table information remaining in the states of SplitAssigner and SourceReader.

The current functionality of scanning newly added tables will removing and cleaning up table info in the previous savepoint according to the new configuration.

The changes proposed in this PR is as follows:

In MySqlSplitAssigner, it will apply the new filter in the new configuration and remove redundant table info
In MySqlSourceReader, it will sieve out splits that do not match the new filter.

[zmzeng]

Hi @qidian99, I wonder why you PR didn't merged into master?
Recently, I came across the same issue as you did, thus this PR is pretty necessary for us.

[ruanhang1993]

@leonardBang Do you have time to review this PR?

[ruanhang1993]

Hi, @qidian99 . Could you rebase the master branch? Thanks~

[leonardBang]

Thanks @qidian99 and @ruanhang1993 for the contribution, I left some coments

[leonardBang]

Thanks @ruanhang1993 for the update, LGTM

[zhangkunjie]

1、I configured MySQL tables A and B to synchronize to a pipeline in Starrocks using the configuration file yaml Task.
2、I use the flink savepoint command to save the running task as a _metadata.
3、I modified the yaml configuration file by commenting out Table B, leaving only Table A, and then restoring the task from the second step's savepoint. At this point, the synchronization task only synchronizes Table A, which is not a problem.
4、At this point, I used the flink savepoint command again to save the running task as a _metadata. Through observation, it was found that the information of Table B, which had already been deleted, still exists in the _metadata at this time. Why does the deleted B table information still exist in the _metadata?