Brainstorm and refactor the discovery process and address book

[tomaka]

There exists multiple systems that are spread between the network services and the network/service:

• "Out slots" are assigned by the network services following the suggestions of the network/service.
• Kbuckets, used to determine which PeerIds to hold (prevents memory explosion) and answer Kademlia requests.
• For each peer, a list of addresses and whether we are connecting or connected to it.
• A ban system in network services, where we prevent connections to the same peer for 20 seconds after a disconnect.

This is in general overly complicated and hard to prove correct.
Additionally, the network services don't have enough control over this process.

The whole thing should be clarified and simplified in the code.