Race in mysql-test/suite/rpl/t/rpl_gtid_stop_start.test

There's a race in mysql-test/suite/rpl/t/rpl_gtid_stop_start.test as follows:

The rpl_gtid_stop_start test used SELECT COUNT(*) as a wait_condition when waiting for rows to appear from the replication stream. Since the table it is waiting on is MyISAM, rows appearing there do not guarantee in any way that the GTID information will have been updated by the slave thread (and can't guarantee that). This causes a race in the test which is especially exacerbated by making the gtid_slave_pos table InnoDB, since then the COMMIT that is done as part of recording the GTID is bound to take some tens of milliseconds.