Changed lots of peer use, and changed the peerstore to ensure
there is only ever one peer in use.

Fixed #174

![](http://m.memegen.com/77n7dk.jpg)

Now, all peers should be retrieved from the Peerstore, which will
construct the peers accordingly. This ensures there's only one peer
object per peer (opposite would be bad: things get out sync)

cc @whyrusleeping

object plumbing commands

Some fixes to bitswap + dht

need to switch away from timing tests. its very annoying.
Should use proper go sync, as in
https://www.youtube.com/watch?v=ndmB0bj7eyw

for some reason travis has more goroutines running by def.

important to always reread your code.

@whyrusleeping pre-commit hook?

It's a patch, really. it's not the full multiconn fix.

- does end properly
- no goroutines leaked!

This commit actually removes the previously introduced
chan net.NetMessage, in favor of raw []byte. It plays
nicer with crypto/spipe, and it makes more sense in the
context of a "single connection", i.e. I already know the
peer I'm talking to, from the connection. The NetMessage
additional Peer is useful swarm and up.