[9.0] fix (WMS): clear the signal mask before starting the payload

[chrisburr]

Creating a gfal2 context (gfal2.creat_context()) leaves most signals blocked in the calling thread (SigBlk: fffffffe3ff9ea07: SIGHUP, SIGINT, SIGQUIT, SIGUSR1/2, SIGALRM, SIGTERM, ...). The signal mask is inherited across fork() and kept by exec(), so every payload started by the JobWrapper after a gfal2 context was created runs with these signals blocked:

• the payload ignores SIGTERM, e.g. from the batch system before a hard kill;
• Gaudi's EventWatchdogAlg can't abort a stuck event, because it uses raise(SIGQUIT): in LHCb productions it logged FATAL ... aborting process 17 times over ~27 hours without the job stopping (gaudi/Gaudi#428);
• Ctrl-C doesn't stop a payload when running a job locally.

This clears the signal mask in ExecutionThread just before it starts the payload. The thread only exists to run the payload, so the rest of the JobWrapper is unaffected, and no preexec_fn is needed.

BEGINRELEASENOTES
*WorkloadManagement
FIX: the payload no longer inherits signals blocked in the JobWrapper (e.g. by gfal2), so it can be stopped with SIGTERM and abort itself with SIGQUIT
ENDRELEASENOTES