sporadically".

Races in truncate_coverage.test have caused its sporadical 
failures.

In the test case we have tried to kill truncate statement 
being executed in the first connection which was waiting 
for X metadata lock on table being locked by the second
connection. Since we have released metadata lock held by 
the second connection right after issuing KILL statement 
sometimes TRUNCATE TABLE managed to acquire X lock before 
it has noticed that it was killed. In this case TRUNCATE
TABLE was successfully executed till its end and this fact
has caused test failure since this statement didn't return 
expected error in such case.

This patch addresses the problem by not releasing metadata
locks in the second connections prematurely.