Optimize local base image support: cache compressed layers

[TadCordle]

Following up #1906. With the initial iteration of local base image support, none of the base image layers are cached. This means every build requires the costly operation of compressing the layers (if not already compressed) and re-calculating the compressed layer digests. Some ideas we can try to get around this:

• Try caching the compressed layers using their diff-ID, or some other easily accessible layer identifier. (Cache compressed local base image layers #1957)
• Cache an entire image to avoid performing docker save every time, and retrieve using the image id reported by docker inspect. (Skip docker save if docker daemon base image is cached #2049)