Dockerfile GHA don't reuse base images

[bbockelm]

I noticed (at least for PR builds) that the build-server-images matrix rebuilds the base image each time.

That is, we currently build the base image and, if successful, we rebuild the base image 8 additional times (one for each derivative of the base image) in parallel. That's around an hour of GitHub action minutes per PR ... which is not so great.

Either pass the images between the build steps or let's remove the parallelism (so we're only building twice!).

[brianaydemir]

Using this 7.13.0-rc.1 action run as a point of reference:

• The build-base-image step is successfully caching the final-stage image.

• The build-base-image step is failing to cache the Pelican binaries because it doesn't depend on pelican-build.

As a result, when the action goes through the build matrix, it ends up building the Pelican binary eight times.

Actually, final-stage does depend on pelican-build. The problem is that pelican-build depends on the entire GitHub workspace (via COPY . .), which includes the .git directory, which I believe contains files with embedded timestamps. Because each job contains its own checkout of the repository, none of them can use the build cache for the Pelican binary.

[brianaydemir]

The other issue is that sometimes, the attempt to cache image layers straight up fails, as in this run:

As a result, everything in the matrix ends up getting built from scratch.