WIP: Use skopeo to fetch images

[cgwalters]

The goal here is to implement: #6

We want to honor things like mirroring set up in /etc/containers.

However, after doing all this work I hit on two things:

• containers/image barfs if the destination is a pipe
• And actually even if we ask to copy into an oci-archive it still
  spools everything to a temporary directory and then tars it
  back up, entirely obviating the point of streaming.

[cgwalters]

For me also this was a huge deep dive into async Rust. I would say I'm intermediate at Rust but I'd only dabbled in async.

In particular the parse_tar function is "bridging" the world of async rust to sync because I wanted to keep using the existing sync tar crate. So I spent a lot of time trying to figure out how to take an AsyncRead and turn that into a plain Read usable in a worker thread.

Then further, I wanted to stream output too, so I ended up using tokio's mpsc channels which are documented to be usable both sync and async. But even this took me a bit to figure out from searching through the tokio docs. Web searches for "async sync bridge read" and the like aren't helpful - the terms are too generic.

Further complicating things I wanted to return an impl Stream<Item=Result<TarEntry>> but we also need to poll the future for the helpers. I tried making my own struct for this but implementing poll_next() was a whole other level of learning - particularly around pinning the inner futures, etc. I ended up just returning two futures.

In addition to this I also spent a lot of time on inscrutable compiler errors that result when e.g. you forget to include TryStreamExt or you forget to wrap the result in a future in a helper closure, etc. Also some lifetime issues that I am still not sure I quite understand (e.g. my use of the Rc<String>), and I ended up tossing in various .boxed_local() for lifetime issues on some of the futures.

Now of course there is https://crates.io/crates/tokio-tar but in a very Rust thing, there's an open PR to port to tokio 1.0 that hasn't merged because it's probably just someone's side project or something.

And....after doing all that I very belatedly realized that skopeo doesn't actually stream the results at all 😢 - so all this work on parsing the tar stream is pointless, I could have just forked skopeo to write to a tempdir, and then parse that.

Ultimately I think we need a more direct containers/image backend to make this truly work well.