Seconds_behind_master incorrect in parallel replication

As reported by plinux on #maria:

09:03 < plinux> We found a display bug about parallel replication. When we update one row on master, and then stop
the updates on master, we can found Seconds_Behind_Master is still increasing on slave

I think the problem is a race between the SQL driver thread and the worker
threads.

Normally, the rli->last_master_timestamp is updated in stmt_done() to the
timestamp of an event from the master. When next_event() reaches the end of
the relay log, it is temporarily set to 0 to mark that SQL thread is caught
up; SHOW SLAVE STATUS handles this value special.

But with parallel replication, it can easily happen that the SQL driver thread
reaches the end of the relay log before a worker thread calls stmt_done(),
thus ending up with the wrong value in rli->last_master_timestamp and hence
wrong output of SHOW SLAVE STATUS.