utils/torture.c: Added one more delete on close test.

utils/torture.c: Added one more delete on close test.
smbd/notify_kernel.c: This code was wrong I believe. It was structured to only
return a changenotify event on being called from timeout processing (t != 0).
The kernel changenotify events should fire on *asynchronous* processing (EINTR
return from select caused by the realtime signal delivery) with t == 0.
Reported by Juergen Hasch (Hasch@t-online.de).
ANDREW PLEASE CHECK THIS !
Currently the hash style changenotify is done on async processing as well
as timeout processing. As this is expensive we may want to revisit doing this
and maybe set it to fire only on timeout processing.
Jeremy.
(This used to be commit f952380c5d0cfbbec1e7de98b712952f302cddfd)