Set up a workflow to build a PETSc Docker image for faster CI builds

[jeff-cohere]

Our CI builds are slow, and most of the time is spent building PETSc. It would be nice if we could use a TDycore-specific Docker image with a version of PETSc (identified by a git hash). If no Docker image existed for a requested hash, a workflow could be triggered that built and pushed such an image to GitHub's Docker registry, and then we could do our CI build on top.

In short, we need

• A Dockerfile that can build PETSc given a specific hash
• A workflow that uses the Dockerfile to build PETSc with that hash, and pushes it to GitHub's container registry with a tag that includes the hash
• A way to specify the hash. It looks like it might be easiest to specify a hash in the PETSc build workflow and have it write out an artifact (a file containing the given hash) that can be picked up and used by the PR auto-test workflow to find the right Docker image.

I'm grafting together a couple of Jed's Docker files to build MPICH from source and install a specific PETSc hash on top of it.

This Dockerfile can then be used in a workflow that builds and pushes Docker images: https://github.com/marketplace/actions/build-and-push-docker-images#cache-to-registry

More resources:

• Storing workflow data as artifacts
• Setting environment variables from within GitHub Actions
• Migrating to GitHub Container Registry

[jeff-cohere]

The workflows described in the above link require a Docker image stored either in DockerHub or in the GitHub Container Registry. The latter seems like it could be simpler, since it's integrated with our organization. Unfortunately, it's in beta right now, meaning there are still some issues to be shaken out. The documentation is clearly not ready for primetime at the moment. I'll keep my eye on this and see how things develop.