Embed Data Dependency Information in Binlog Events

The current parallelization mechanism for the parallel replica (i.e. slave-parallel-mode=optimistic/aggressive) relies only on commit-ordering constraints (exposed via GTID) to be able to safely parallelize the execution of transactions. That is, the parallel replica will try to run transactions concurrently, and then if there is a conflict, the later transaction is rolled back and re-executed from the beginning once the prior transaction has commit. This leads to additional time & energy spent to roll-back any conflicting transactions and re-execute the same work which had to be rolled back.

This loss of time/energy can be improved by embedding dependency tracking information into binlog events, so the transactions can be more intelligently scheduled as to minimize scheduling conflicting transactions together.