Pin/Unpin peer addresses

[aarshkshah1992]

@raulk

For libp2p/go-libp2p-kad-dht#283.
See theaddress management section in this comment for the rationale.

[raulk]

There are various complex concerns with this solution:

• Peers don’t expire, individual addresses expire.
• Address decay is how we maintain an ever-changing set of addresses for a peer.
• Addresses not reported/observed recently will expire, while the recent ones will be bumped up.
• Does pinning prevent all addresses from expiring? Then we could be left with stale addresses, as the above decay wouldn’t have had a chance to expire them.
• If anything, this should be “best effort”. We can’t afford for buggy callers that do not call Unpin properly to cause garbage to accumulate.
• Pin/Unpin would need to do reference counting. What if two components call Pin, and only one calls Unpin?

[aarshkshah1992]

We have an issue open to rethink our peerstore design to solve nasty problems such as this one:

libp2p/specs#164

[jacobheun]

I'm going to close this since it's from February.