Make TIER2 connections survive node restart

[saketh-are]

This PR implements stabilization for the TIER2 network:

• Adds ConnectionStore, which maintains an LRU cache of outbound connections and persists them to the database.
• Adds behavior in PeerManagerActor to re-establish stored outbound connections upon starting a node.
• Adds behavior in PeerManagerActor to re-establish an active outbound connection if it is in the connection store and the connection is broken.
• Adds ConnectionStore data to the TIER2 network debug page.

A flag remove_from_connection_store is added to Disconnect messages. The flag is set to true by the disconnecting node if it does not expect to accept reconnection from the peer, advising the peer to remove the connection from its connection store if present. If a peer sends an old-version Disconnect message without the field, remove_from_... defaults to false and the connection will be re-attempted if it is in the ConnectionStore.

TODOs:

• Add a command line flag such as --ignore-rc to skip reconnection when starting a node

[saketh-are]

As before, ready for another look apart from renaming remove_from_recent_outbound_connections everywhere.

[pompon0]

It looks very good now! Just one test needs fixing - the one with the clock rewinding (I called it clock skew in the previous comment, but skew is sth different, sorry), but other than that you can go ahead and merge. The flag can be added in a separate PR. I think that we can name the flag to sth like:
"--connect_to_reliable_peers_on_startup" with default "true".