MEMORY TABLE
Bug#25681518: CLEANUP PROCEDURE FOR MEMORY ENGINE TABLES
              BREAKS GTID REPLICA

Analysis: Memory table's data will be lost after server
restarts. To keep the data consistency between master and
slave, it just binlogs an implicit DELETE statement when
the memory table is locked first time.

When the above scenario happens in a replication thread,
the server raises an assertion. The reason is that the
implicit DELETE statement consumes the GTID of the actual
transaction (the one that is stored in the relay log), so
that the the actual transaction, which comes later, has no
GTID to use. This causes the assertion.

Fix: This issue is fixed by creating a new THD object for
handling the implicit binary logging of DELETE statement.
This ensures that the implicit DELETE as well as the actual
trasaction which triggered the implicit DELETE gets binary
logged with the correct GTIDs. This fix also ensures that
the implicit DELETE will be binlogged separately and will
be wrapped by BEGIN/COMMIT pair.

RB#24351