GitLab

This is a huge performance hit. Really, we just need to tell users not to modify
the result.

Also, get rid of an unnecessary pointer indirection (no api change).

fixes part one of #77

See [rfc4007].

Can be prefixed to "ip6" like so:

    /zone/eth0/ip6/fe80::1

This corresponds to the rfc4007 address `fe80::1%eth0`

[rfc4007] https://tools.ietf.org/html/rfc4007

It conflicted with CBOR.

part of multiformats/multicodec#55

This commit changes the struct to a new Multiaddr interface:

```Go
type Multiaddr interface {
  Equal(Multiaddr) bool
  Bytes() []byte
  String() string
  Protocols() []*Protocol
  Encapsulate(Multiaddr) Multiaddr
  Decapsulate(Multiaddr) Multiaddr
}
```

This means a few things have changed:

- use Multiaddr interface, struct not exported
- Bytes returns a copy of the internal bytes
- Some methods no longer return errors (catch errors in NewMultiaddr)
  - String (panics if malformed)
  - Protocols (panics if malformed)
  - Decapsulate (no-op if not prefix)
- Moved net-specific functions to package
  - Multiaddr.DialArgs() -> DialArgs(Multiaddr)
  - Multiaddr.IsThinWaist() -> IsThinWaist(Multiaddr)

cc @whyrusleeping @perfmode

The error in String should not actually ocurr, as the multiaddr
should have been valid to be constructed successfully, and thus
should be encoded back to its string rep correctly.

This will be bolstered by creating an interface (to prevent messing
with the internal bytes)