The query present in the sql dump sets gtid_slave_pos, which is used by
slave_pos. current_pos relies on gtid_current_pos, which is not set here
and hence fails to initiate replication.
Another (unintended) effect of using slave_pos is that queries run on the
slave will not break replication. There should be no reason to run queries
on slave (at least, data & schema modification queries while replication
is active), so it would seem better to fail the replication immediately in
order to detect this. So this may not be the best solution - but at least
this fixes this script.