We cannot sleep with the activity mutex held without affecting
parallelism so badly. Otoh, we cannot wleep on UpdatesNotificationResourceLock
without affecting 3rd party clients in a way that we would starve them
for a while. So, try to acquire the file lock in non blocking mode
and if it fails, reschedule a new call as soon as the same lock is
freed (exclusive acquire succeeds). This assumes that if the lock
is held back, we will just re-do the whole trick again and again.