NDBCLUSTER GLOBAL SCHEMA LOCK

When running a query on I_S.TABLES an endless loop is entered while
waiting for global schema lock to be available. It's not possible to
kill the query, have the query "kill itself" using the
MAX_EXECUTION_TIME optimizer hint nor using --max-execution--time
variable.

Problem occurs when the attempt to acquire the GSL row-lock failed due
to do deadlock detection timeout error in NDB. The query is then choosen
as "victimized", causing endless retries without checking the killed
flag. Both DDL and `SELECT ... FROM I_S.TABLES` queries are affected.

Fix by checking the killed flag and terminate the retry loop.

Also add an optimisation to ndbcluster_find_files() making it check the
killed flag and return immediately, this significantly reduces total time
for I_S.TABLES queries to terminate.

Also fix a problem with calling unlock after error which causes variable
underflow. Theoretically it should not be necessary to call unlock when
lock failed, but it's currently not possible to eliminate the unlock
call.

Add tests for global schema lock in combination with SELECT I_S.TABLES
query using MAX_EXECUTION_TIME as well as DDL query being killed while
waiting for global schema lock.

Change-Id: If9e5c9787c347c400a66d2edf00cc59abe993419