Consider resetting wfu/mtbf/tk values for relays when they switch IP addresses

The proposal would be to track the IP address of each relay in the router-stability file, and when the IP address changes for a given relay identity, start fresh on tracking the weighted-fractional-uptime, mean-time-between-failure, and time-known.

The benefit is that if some bad person seizes a relay and gets the identity key, then puts that identity key online somewhere else, clients won't immediately resume using it as their guard. We would have at least a week to notice that it's back. This feature in turn would make it less critical to blacklist identity fingerprints of seized relays, e.g. https://lists.torproject.org/pipermail/tor-relays/2017-May/012319.html

The drawback is that we make less good use of relays on dynamic IP addresses, since they will spend a lot of their time not having the Guard or Stable flags.

Note that this change wouldn't impact the bandwidth authority numbers, so it wouldn't make relays need to wait until the bwauths have measured them before getting traffic again.