Graphsync Transfer Failures

I'm noticing a lot of graphsync transfer failures for a number of different reasons (known and unknown).

[ ] The first issue is that yamux will periodically time out and close the whole connection, this is actually a bug and should be resolved in https://github.com/libp2p/go-yamux/pull/35
[ ] data transfers appear to fail rather often when restarted with 'message too large'
[ ] data transfer doesnt appear to automatically restart anything (we probably should, some limited number of times)
[ ] counting of data sent and received appears to be incorrect around restarts
[ ] data transfers appear to fail due to inactivity somewhat often, for inexplicable reasons
[ ] data transfers get removed due to some timeout, but it appears to be different between client and server. I think I had a transfer get removed by the miner side while the client was still transferring data

filecoin-project / lotus