Optimize CI actions to use fewer minutes (/w less parallelization or more caching)

[NevilleS]

Right now our CI has been very helpful, but as we continue to add more tests some of the inherent inefficiencies are really adding up!

After #86 lands our checks will look something like this:

If I'm following correctly, I think that's 8 jobs, totaling 33 minutes of CI actions for one PR 😅

Of course our actual actions (tests, linters, etc.) are quite fast - but the slow part is that each of these runners needs to docker pull all the dependencies and build the images every time. There's no caching between actions either so we pay that tax on every job.

Options here are:

1. Combine multiple jobs into a single job, so that we can reuse the Docker images across them
2. Setup Docker caching in our CI (I think this looks right: https://evilmartians.com/chronicles/build-images-on-github-actions-with-docker-layer-caching)

[NevilleS]

After some googling and reading of issue comment threads, I came across this example for enabling caching: docker/build-push-action#406 (comment)

However, we can't use that directly as our docker build commands are in the makefile and have other customizations, etc. etc. Effectively we'd need to recreate some of that cache layer in how we invoke the docker commands and it doesn't seem like that'd be particularly easy to do in a transparent way...