filter unworkable peers in queries + enhanced logging

[raulk]

The critical contribution of this PR is preventing adding useless peers to our queries. The definition of useless is:

• peers with no addresses.
• peers with relay addresses (therefore unupgraded DHT clients).
• peers with no publicly routable addresses (unless we're connected to the peer via LAN).

Secondly it also adds outbound filtering so that upgraded nodes will contextualise the peers they send back, based on whether the connection with the requester is via LAN or not.

[Stebalien]

We need to put/get to a consistent set of peers. This current patch doesn't do that. It may work in practice because, unless we're entirely offline, we're only likely to find at most one "closest peer" on the local network (out of K=20) but it's still incorrect and this at least needs to be documented somewhere.

We could also:

1. Have a disconnected/connected "switch": when we add the first non-local peer to our routing table, we remove all unroutable peers, when we run out of routable peers, we rebuild our routing table from unroutable peers. This won't break local peer routing as we return the target of FindPeer even if it isn't in our routing table.
2. Make sure to return at least 20 global peers if we have them, even if we have 20 closer local peers. Also make sure to put to 20 routable peers. This solution is closest to the existing solution but really hard to get right.
3. Run two DHTs (overhead should be minimal as the "local" DHT should be small). This is probably the most elegant solution.