This commit makes all network tests use ZeroLocalTCPAddress
as the initial peer address, and then relies on
net.ListenAddresses()

This should get rid of the tcp addr clash problems.

The separation of work in the p2p pkg is as follows:

- net implements the Swarm and connectivity
- protocol has muxer and header protocols
- host implements protocol muxing + services
- identify took over handshake completely! yay.
- p2p package works as a whole

using a placeholder net2 package so tests continue to pass.
Will be swapped atomically into main code.

remove comment about spdystream. this is now
superceded by go-peerstream interface.

The net package is the next to move. It will be massaged
a bit still to fix the Network / "NetworkBackend" conflict.

The crypto package moves into p2p. Nothing in it so far is ipfs
specific; everything is p2p-general.

I think it's time to move a lot of the peer-to-peer networking
but-not-ipfs-specific things into its own package: p2p.
This could in the future be split off into its own library.
The first thing to go is the peer.

These services should all be added separately, in a function
that is Network impl independent. The interfaces need to be
massaged a bit (split inet.Network in two), so will do when
needed. For now this is fine.

swarmnet is a better name for the package, because
it's just a Network implemented with a Swarm.
(ipfsnet will be something slightly different).

this is the leadup into NAT traversal.
note: doesn't work yet. hangs the test.

I needed the network implementation in its own
package, because I'll be writing several services that
will plug into _it_ that shouldn't be part of the core net
package. and then there were dependency conflicts. yay.
mux + identify are good examples of what i mean.

blockstore: suppress exchange error

rewrite of provides to better select peers to send RPCs to

Also remove older useless code.

The test was occasionally passing because:
- it called `putLocal(key, val)`
- GetValue calls `getLocal(key)` optimistically.

cc @whyrusleeping

refactor test peer creation to be deterministic and reliable

a bit of cleanup trying to figure out TestGetFailure

add test to verify deterministic peer creation

switch put RPC over to use getClosestPeers

rm 0xDEADC0DE

fix queries not searching peer if its not actually closer

peerstream transports

go-peerstream update to use github.com/hashicorp/yamux

clients were already accessing secio in one thread, but it's
safer to make sure it _is_ threadsafe

used to return io.ErrShortBuffer, but this makes client
code much more complicated. we're already allocating
buffers when it's too large, so might as well just
keep it for later.

peerstream was updated to use pluggable transports, including
muxado. The interface was also simplified slightly.

bugfix in spdystream