Optional CI job that runs the GPU tests on a simulated GPU (PantheonSim) #2

The CI here builds the docs and checks links, but the CUDA code paths are never run, because the hosted runners have no GPU.

I am the author of PantheonSim, a functional GPU simulator that runs CUDA programs without a GPU, and of the GitHub Action pantheongpu/setup-pantheonsim that sets it up on a hosted runner. I would like to propose a separate, optional workflow that builds DYNAMITE with CUDA and runs the GPU tests on a simulated H100. It would not replace testing on real GPUs, because the simulator has no timing model and cannot reproduce hardware-specific behaviour.

I have not tried it on DYNAMITE yet. If you are open to it, I will first check on my own fork that it builds and which tests pass, and only then send a small pull request (the workflow file only), linked to this issue. For reference, here is a pull request of the same kind on another project: CERN/TIGRE#808

Would you want this? I am the author of PantheonSim, so please weigh this accordingly.