[ci/cd] Optimize docker build

[mdartic]

• parallelize with matrix
• use cache with cache-form / cache-to

Here are some links :
https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners
https://docs.docker.com/engine/reference/commandline/buildx_build/#cache-to
https://github.com/moby/buildkit#export-cache
https://docs.github.com/en/actions/using-jobs/using-a-matrix-for-your-jobs

[mdartic]

So, after discussion with @kiorky, I try several things :

• parallelization with matrix on platforms
  • an issue appears, with the build & push github action, the latest build is pushed and overwrite all previously images
  • a better idea of this could be the use of docker manifest to create a multi-arch manifest (https://www.docker.com/blog/multi-arch-build-and-images-the-simple-way/) AFTER all platforms are built
  • this could also help us if we want to test all built images (particularly arm32 ones)
  • the last step of the workflow could be the push of the tag with the multi-arch manifest
• set the cache-from / cache-to
  • I don't see any improvements on build time
  • I let them for now and check if there is some benefits
• move the docker build job in the api-quality workflow to reuse build artifacts
  • this is the main shift, as it cut off the build time from one hour to ~6mn
  • and this, without the use of matrix as it breaks the push of multi-arch images

Initial time:

https://github.com/locokit/locokit/actions/runs/5923472268

After use of parallelization with matrix on platforms:

https://github.com/locokit/locokit/actions/runs/5965675095

After shifting docker build in the api quality workflow:

with parallelization (and issue when pushing tag to hub.docker.com)

https://github.com/locokit/locokit/actions/runs/5966002861

without parallelization (and no issue on push)

https://github.com/locokit/locokit/actions/runs/5966049744

For now, we'll stay on what is in the PR, so :

• shift of the docker build in the api quality workflow
• reuse of build artifacts for docker build
• no matrix / parallelization by platform