Reconnecting an existing socket simply was not working; the Rust socket
did not expose any methods for reconnection, so I simply tried calling
connect() again. As near as I can tell, this was a no-op --- which makes
sense because both the sending and receiving threads had their own copy
of the Socket, and it's not clear what the synchronization behaviour
should have been.
Instead if the connection fails, we relay this information to the main
thread, wait for an acknowledgement, then simply destroy the listening
thread. The caller can then simply call `start()` again.
This is a massive simplification, fixes a couple endianness bugs (though
not all of them I don't think), should give a speedup, gets rid of the
`serialize_iter` crap.
It is now visible that EOF (i.e. peer hung up) is interpreted
as a message decode error. Probably what we want to do is reset
the connection on any error. TODO
Andrew Poelstra