P2P: Fix duplicate connection logic

[heifner]

See #1243 for additional info.

The current scheme still leaves duplicates in some race conditions. A simpler approach was recently attempted. See #1108. However, this approach had a race condition so that it would sometimes disconnect both connections; reverted in #1186. The issue is that the connection and handshake messages can be sent from each node at the same time. Both nodes can consider the other a duplicate at the same time and both disconnect.

We believe the race condition can be solved using the following technique.

1. On disconnect because of duplicate connection, do not attempt an immediate retry of the connections.
2. When connecting to a peer, verify there is not already a open connection to that peer node_id.
   a. Note the node_id of the connection must be saved on close to make this determination.
3. Maintain the logic of closing a duplicate if node's node_id less than peer node_id.

[heifner]

1. Already was not attempting to immediately retry on duplicate disconnect.
2. In order to know the node_id a handshake has to be received from the connection. This can be accomplished by checking both incoming and outgoing initial handshakes for duplicate.
3. Nothing to do.