Adjust LogMiner mining algorithm to reduce or avoid the need to re-mine [DBZ-3654]

Migrated from [DBZ-3654](https://issues.redhat.com/browse/DBZ-3654)

When BLOB/CLOB support was introduced to the Oracle connector, the streaming logic needed to shift from using a sliding adaptive mining window where the next iteration began where the prior left off, to using a sticky starting point. This means that the connector will potentially continually re-mine old records until the oldest active transaction concludes with either a commit or rollback. For high volume systems, this can potentially be expensive and so trying to find a way to reduce the cost of the sticky starting point approach or getting back to the start where we left off concept would be ideal.