Fix: fencer: prevent watchdog fencing targeting an offline node from timing out on insufficient stonith-timeout

A specified/defaulted global stonith-timeout shorter than
stonith-watchdog-timeout could be arguably considered a
misconfiguration. But given the configured stonith-watchdog-timeout,
it's pretty well known how long a watchdog fencing takes. In that case,
we should ideally tolerate the misconfiguration and allow watchdog
fencing to take as much time as it needs.

In any case, stonith-timeout still matters at least as the query
timeout.