Dockerfile for CPU-optimized Sockeye image

[mjdenkowski]

This commit adds a Dockerfile and build script for an experimental CPU-optimized Sockeye image. The resulting image includes the following:

• Intel MKL 2019.4 downloaded from the apt repository.
• kpuatamazon/incubator-mxnet: The MXNet fork that supports intgemm and makes full use of Intel MKL (versus just DNNL).
• kpuatamazon/sockeye: The Sockeye fork that supports int8 quantization for inference.

The updated docker/README.md includes instructions for using this image.

This requires no code changes to the sockeye_2 branch (version is only incremented for tracking changes).

Pull Request Checklist

• Changes are complete (if posting work-in-progress code, prefix your pull request title with '[WIP]'
  until you can check this box.
• Updated major/minor version in sockeye/__init__.py. Major version bump if this is a backwards incompatible change.
• Updated CHANGELOG.md

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.