fix: one peer occupy multiple tcp connections

[garyyu]

To close #2258

• The main improvement is closing TCP stream on fail of peer connect / accept
• Another correction, (not directly related to this issue), is removing the unnecessary 3 times retry for peer connection request, which was imported by me before, but I think I was over worried on the connection request, since we have 10s timeout for a TCP connecting request:

TcpStream::connect_timeout(addr, Duration::from_secs(10))

And it doesn't make sense to retry in 1 second on any error, and we already have a resurrection mechanism for those p2p::State::Defunct peers.

[garyyu]

Tested in 3 servers for 8 hours, the effect is quite good, no more hundreds of connections.

But for 100% avoid multiple tcp connections on one peer, we need more actions:
1.Ensure all grin nodes have been upgraded to this version (or later versions)
2.New mechanism to save all connected TCP streams, and periodically check and clean those connections which has no data flow (protocol data sending / receiving). This deserves a new PR to do that, but perhaps do it later since it's not a major issue now.