Define CUDA versions for workflows centrally

[rickardp]

This PR defines CUDA versions to be built in workflows centrally.

It uses a Dockerfile as a source of truth, as this can be updated automatically by Dependabot.

Currently, the Dockerfile is not used for building the compiler images, but we could do this in the future if we want to speed up buiilds (by caching on ghcr.io and e.g stripping static libs that we don't use, as they constitute about 50% of the image size), but the bottleneck is Windows anyway so this is not critical at the moment

[github-actions[bot]]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.