the u.Hash error can be safely ignored (panic) because multihash
only fails from the selection of hash function. If the fn + length
are valid, it won't error.
cc @whyrusleeping
This commit fixed the notoriously annoying "Malformed Public Key"
problem. The issue was that sometimes the byte representation of
the points (x,y in big.Int) generated would be one less byte than
expected. This is simply because (* big.Int) Write uses the least
amount of bytes needed for the int.
I instead changed the marshalling/unmarshalling to do exactly
what stdlib crypto/tls does: use `ellipctic.Marshal` which marshals
according to the ANSI X9.62 standard.
http://golang.org/pkg/crypto/elliptic/#Marshalhttp://golang.org/src/pkg/crypto/tls/key_agreement.go#L214
```Go
// crypto/tls
ka.privateKey, x, y, err = elliptic.GenerateKey(ka.curve, config.rand())
ecdhePublic := elliptic.Marshal(ka.curve, x, y)
// ipfs/crypto
priv, x, y, err := elliptic.GenerateKey(curve, rand.Reader)
pubKey := elliptic.Marshal(curve, x, y)
```
((Warning: we're using `rand.Reader` directly, which we shouldn't
do, as it can be seeded. We should use a configured source, as
crypto/tls. Flagged in #143))
This makes me think we should re-use a lot of their datastructures
and functions directly (e.g. ecdheKeyAgreement)
Fixed: #135
cc @bren2010 @whyrusleeping
