Cherry pick of the server part of the patch from mysql-5.7
In adition this commit adds the Cluster specific part of patch

NdbCluster code for acquiring a Global Schema Lock has been changed
to not retry the GSL lock if it failed due to a timeout (default 3000ms),
*and* there is a potential for this lock request participating in a MDL-GSL
deadlock. In these cases it select itself as a 'victim', and return
this decission to the MDL requestor.

MDL code will then either handle such a victimized GSL request either as
a failed lock request - Which is better than staying in a deadlock state
forever.

Or, where a deadlock_handler exist in the MDL code, it will backof
and retry the MDL+GSL locking again. (Valid for CREATE TABLE which
was the soure for the deadlocks in the attached test case)