We think it is more likely that nodes accidentally are offline for two
hours and then would switch prematurely, than that they have incorrect
system time during switchover and then switch one day too late.
- add some of our own VPN servers, so that the check also works for nodes
that may only connect to the VPN servers
- remove one of the Google servers, to keep the number of servers low (the
ping check is executed every minute and sends one ping packet to each IP,
with a wait time of 10 seconds; so let's keep the total check time under
60 seconds)