Changes proposed in this PR

This implements the feature I suggested a few years ago here: #167

Basically, I need to extract only a portion of length n of a compressed file, at a particular offset i. The use case is serving the entry over an HTTP byte-range request, which should not be entirely loaded into memory.

In the case of stored entries, I simply forward the requested range without modification.

Tests performed

I'm already using this implementation in our toolkit (Readium). I can write unit tests or do modifications if you're interested in this feature @weichsel.

Further info for the reviewer

I implemented other features in our fork that I would love to get merged upstream, if that's of interest to you.

• Isolated the data access layer behind a DataSource protocol, that we used to stream ZIP entries over HTTP without downloading them.
• Made the APIs async.