Rather than taking in a Stream and advancing it internally, return a
Sink that can be advanced by the calling code. This significantly
simplifies encoding logic for things like tokio-postgres-binary-copy.
Similarly, the blocking interface returns a Writer.
Closes#489
Benchmarks indicate that malloc accounts for a significant amount of the
runtime of queries. The message buffer accounts for ~half of that (the
other being channels), and this change should eliminate it.
