Cache results of hooks

[JonathonReinhart]

When using Scuba to emulate a CI runner, the hooks may perform expensive operations. For example, a .gitlab-ci.yml before_script may run commands to install packages via apt. Running these hooks every time the user runs scuba would be unacceptable to the user.

One alternative would be to manually perform the installation, and push the resulting image to your image repository.

Instead, Scuba could run the hooks just once, and commit the resulting container to an image, which Scuba will use for all subsequent invocations.

This is no small task; caching is difficult. However, it would behave similar to the way docker build caches intermediate layers. We could hash the hook scripts, and use that as a key to refer to the cached image.

Additional difficulties might include things that are done by the hooks that don't affect the filesystem. I'm not sure what this might be, but these hooks would still need executed, even if we're running a cached image. This might require adding an additional flag to the hooks, or adding a separate top-level key.