The connection_timeout value is about just establishing connection. It's
rather short by default (5s), which is normally fine. But there are
cases where it needs to be longer, for example when VM is resuming from
sleep.

QubesOS/qubes-issues#9942

The VM may still be resuming from suspend (for example kernel may still
be in progress of re-initializing grant tables), so the default 5s
sometimes is too little. Increase the connection timeout to the full
'suspend_timeout' - quite excessive, but the whole call is under a
timeout too, so it simply makes different type of timeouts handled
uniformly now. Now, the suspend_timeout effectively covers the whole
call, regardless of how much of that time was spent on waiting for the
connection to be established.

Fixes QubesOS/qubes-issues#9942