Automated binlog purge caused a MySQLD crash.

This was because purging a Binlog file has an associated DELETE 
from the ndb_binlog_index file to remove entries for epochs stored
in the Binlog Index file.

This DELETE is executed using the current 'THD' structure in the 
purge code context.

However, the purge code can be called as a discrete command, in 
which case the THD is quite usable, or just after writing an event
to the Binlog, which can be embedded in the depths of processing
some other statement.  In this case, the THD generally is not in 
a state to be used to start a fresh DELETE command.

To get around this problem, the PURGE DELETE code is changed to 
create and use a fresh THD object.  This THD object is initialised
to the best of the Ndb handler's ability.  Some extra steps are 
taken to make sure it does not itself generate Binlog etc.

A new testcase is added to the ndb_binlog suite to test this 
purge-as-part-of-writing-an-event case.

The testcase uses a new error insertion (DBUG_EXECUTE_IF) added to
Server code.  This code has been pushed to 5.6 and trunk, so is
only an MCP patch up to the 5.6 release.