Fix the testing for JDK 8 /11 by using --build-arg

[sbesson]

Noticed while working on building an image on JDK 17 to start testing ome/bioformats#3796 and ome/bioformats#3815 in our CI

The previous logic was based on environment variables and effectively using the same base image for both matrix builds.
486cddd uses the --build-arg option and also uses the <java_version>-slim-bullseye Docker tag to ensure the image is Debian-based.

To test this build look at the raw logs of the JDK8 and the JDK11 builds. Without this PR, the logs should indicate /usr/local/openjdk-8 in both cases. With this PR the JDK version should match the matrix parameter

[joshmoore]

LGTM. Is your plan to add in 17 to this repo or only consume this downstream?

[sbesson]

LGTM. Is your plan to add in 17 to this repo or only consume this downstream?

😄 Ironically, this branch was original called something like jdk17 and was adding 17 to the matrix until I realised that the build argument logic was broken and also that several low-level components will need some JDK 17 adjustements.

I went for a PR to fix the first issue and I'll discuss the second one this afternoon at the weekly Formats meetings.

[sbesson]

Merging as this only affect the GitHub Actions builds. Next steps will be to:

• update the OME Jenkins infrastructure to start alternating between JDK versions (8 and 11 initially and planning to add 17 eventually)
• start reviewing the low-level Java components, adding 17 to the matrix GitHub builds and fixing build issues as they arise

[sbesson]

See ome/devspace#194 for the update of the Jenkins infrastructure to rotate between JDK versions.