Speaking with the devexp committee, and in the recent community hall
meeting, it's pretty clear we don't need to build the whole docker
matrix on every PR event. It's somewhat expensive and isn't useful.

One of the reason we did it, from what I've gathered, is we want to
catch if some change breaks the docker build for whatever reason. Now
we may not need to actively prevent this pre-merge as this is unlikely,
yet we can still catch it on `master` as I'm keeping the push events on
master.

Note that:
- I'm also moving the master push build to be multi-platform (amd+arm)
  as we started doing this for the release builds, which is nice as
  the same image can be used on different archs
- The PR events were not "pushed" or cached in our shared, remote
  dockerhub-hosted repo anyways, they were just kind of dry-runs

Oh one more thing, notice the no-op.yml changes? That's because the
required checks related to branch protection are always based on what's
defined on .asf.yml ON `master`, meaning i need to add dummy/no-op
checks to get this merged, and those can be cleaned up later on in a
follow up PR. Let me know if you know a better way to deal with this.