Add get_ref and get_mut methods to Decoder and Encoder

chore: Re-license to dual MIT/Apache 2.0

Fix crates.io badge and update links

Expose Decoder::remaining_chunks_size

This is a workaround for using Decoder with tokio::io::AsyncRead. The
line containing the chunk size can be read asynchronously and decoded by
Decoder. Then the chunk itself can be read asynchronously now that its
size is known.

Performance: One write syscall per chunk.

Co-authored-by: Corey Farwell <coreyf@rwell.org>

This ports over an optimization from
https://github.com/algesten/ureq/pull/44, to do just one write syscall
per chunk. It does so efficiently by always reserving space for the
chunk size at the beginning of the buffer, so we don't have to copy lots
of memory around when it comes time to write the chunk.

Update documentation link to point to docs.rs.

The current documentation on github.io was out-of-date. This seemed like
a quick fix to point to the most up-to-date docs.

Add option to flush after each write to encoder

Add Decoder::into_inner method to enable unsafe code removal in ureq.

A couple improvements; inlcude hyper unit test