Update to match master version of IoVec (0.2.0?), using
IoVec/IoVecMut instead of &IoVec and &mut IoVec.

The panic happens when `inner.bytes()` returns a slice smaller than the
limit.

Closes #79

I found this significantly improved a
[benchmark](https://gist.github.com/danburkert/34a7d6680d97bc86dca7f396eb8d0abf)
which calls `bytes_mut`, writes 1 byte, and advances the pointer with
`advance_mut` in a pretty tight loop. In particular, it seems to be the
inline annotation on `bytes_mut` which had the most effect. I also took
the opportunity to simplify the bounds checking in advance_mut.

before:

```
test encode_varint_small  ... bench:         540 ns/iter (+/- 85) = 1481 MB/s
```

after:

```
test encode_varint_small  ... bench:         422 ns/iter (+/- 24) = 1895 MB/s
```

As you can see, the variance is also significantly improved.

Interestingly, I tried to change the last statement in `bytes_mut` from

```
&mut slice::from_raw_parts_mut(ptr, cap)[len..]
```

to

```
slice::from_raw_parts_mut(ptr.offset(len as isize), cap - len)
```

but, this caused a very measurable perf regression (almost completely
negating the gains from marking bytes_mut inline).

Also fixes an issue with a line wrap in the middle of an inline code
block.

The `Source` trait was essentially covering the same case as `IntoBuf`,
so remove it.

While technically a breaking change, this should not have any impact due
to:

1) There are no reverse dependencies that currently depend on `bytes`
2) Source was not supposed to be implemented externally
3) IntoBuf provides the same implementations as `Source`

Given these points, the change should be safe to apply.

Add `?Sized` bounds to work for DST objects and also add impls for `Box` as well
as `&mut`

Enables collecting the contents of a `Buf` value into a relevant
concrete buffer implementation.

* `drain_to` and `split_off` take &self and return Bytes.
* `drain_to_mut` and `split_off_mut` take &mut self and return BytesMut

This commit is a significant overhaul of the library in an effort to head
towards a stable API. The rope implementation as well as a number of buffer
implementations have been removed from the library and will live at
https://github.com/carllerche/bytes-more while they incubate.

**Bytes / BytesMut**

`Bytes` is now an atomic ref counted byte slice. As it is contigous, it offers
a richer API than before.

`BytesMut` is a mutable variant. It is safe by ensuring that it is the only
handle to a given byte slice.

**AppendBuf -> ByteBuf**

`AppendBuf` has been replaced by `ByteBuf`. The API is not identical, but is
close enough to be considered a suitable replacement.

**Removed types**

The following types have been removed in favor of living in bytes-more

* RingBuf
* BlockBuf
* `Bytes` as a rope implementation
* ReadExt
* WriteExt

Fixes #46

Not having `unsafe_no_drop_flag` caused some weirdness with optimizing buffers
and bytes. For now, remeove it.

* Get rid of `ByteStr` trait
* `Bytes` is not a concrete type
* Add `BlockBuf`
* Delete lots of cruft
* Performance work