Switch CI to use containers

[LegNeato]

GitHub Actions appears to support running steps in containers:

https://docs.github.com/en/actions/writing-workflows/choosing-where-your-workflow-runs/running-jobs-in-a-container

It's probably good to just reuse the public NVIDIA containers and/or our Dockerfiles in the repo so that everything is installed by nvidia. This will, for example, fix the fact that cudnn is not installed as they have containers with it. We can focus our efforts on the rust-specific stuff.

NVIDIA's supported container tags show they support ubuntu 20 and 24 with the CUDA versions we need.

@adamcavendish already did work creating Dockerfiles in https://github.com/Rust-GPU/Rust-CUDA/tree/main/container. Ideally we would have one set that we use for CI and point people to.

[adamcavendish]

Sure, I'll look into it a bit tomorrow to see whether I can figure it out and make a PR.

After the landing of previous change, I'll also add clang to support bindgen in the containers. Might need some deps to build examples too.

In addition, what should we do to deal with OptiX licenses so we can add that to containers?

[LegNeato]

I'd wait on optix as I haven't read the license yet.

I didn't mean to suggest you should do this, I was totally going to look at it as I know you have that other PR you are iterating on. But if you want to, go ahead! Otherwise I will poke at it once I finish some other stuff.