CR_SERVER_GONE_ERROR). Instead of bailing, sit and loop trying to
reconnect, given that this is known to be a transient error, and we do
not really want our daemons to go belly up during that brief time when
the watchdog is getting it restarted.

This turns out to be nice in my elabinelab experiments, where very
idle daemons have a habit of timing out, although there were other
ways to deal with that, but I'm lazy.