io: remove bytes dependency from the tokio public API

[carllerche]

Summary

The bytes crate is not yet at 1.0. Removing bytes from the Tokio public API will enable us to ship faster. It will also allow both bytes and Tokio to experiment with additional potential APIs before stabilizing. Any utility currently in tokio that uses bytes types as part of the public API is moved to tokio-util.

Motivation

Tokio aims to release 1.0 by the end of 2020. Any type from external crates available as part of the Tokio public API must follow the same guarantees as Toko. The bytes crate is not currently at a 1.0. While we could attempt to reach a bytes 1.0 release, doing so is risky for a few reasons.

• There are still open API questions around exposing the vtable API.
• We may find after experimenting with io_uring integration (io: propose new AsyncRead / AsyncWrite traits #2716) that we need additional / different Buf / BufMut APIs.

Removing bytes as a public dependency for the initial 1.0 release would be relatively straightforward as it is not used as #2716 proposes to remove it from the main core API.

Proposal

All utilities that use bytes are moved to tokio-util until they can be stabilized. This includes:

• AsyncRead::read_buf
• AsyncWrite::write_buf
• Stream::collect() for bytes types.
• stream_reader

AsyncReadExt and AsyncWriteExt traits are added to tokio-util (read / write helpers in tokio are moved to AsyncRead / AsyncWrite proper).

The loss of StreamExt::collect will be the most immediately impactful. We could add StreamExt::collect_buf() (naming TBD) in tokio-util.

[carllerche]

Instead of avoiding bytes, we are going to release bytes 1.0 and depend on that.

Refs: #3058