Add Cuda test workflow for push action in sycl_develop

[aacostadiaz]

This PR introduces a cuda test workflow triggered on push actions in the sycl_develop branch. The workflow will build the entire library which may take a considerable amount of time

[carlewis]

Unless I'm missing some detail, if we are not going to run anything in the device, but only build the code we can take a different approach.

We can use runs-on: ubuntu-22.04 or runs-on: ubuntu-latest, and also use a public docker image with CUDA installed. There are a few different tags available here:

https://hub.docker.com/r/nvidia/cuda/tags

Something like: image: nvidia/gpu:12.8.0-base-ubuntu22.04

This way the GPU runner won't be monopolized by this workflow.

[aacostadiaz]

Unless I'm missing some detail, if we are not going to run anything in the device, but only build the code we can take a different approach.

We can use runs-on: ubuntu-22.04 or runs-on: ubuntu-latest, and also use a public docker image with CUDA installed. There are a few different tags available here:

https://hub.docker.com/r/nvidia/cuda/tags

Something like: image: nvidia/gpu:12.8.0-base-ubuntu22.04

This way the GPU runner won't be monopolized by this workflow.

it is running the unit tests on device too. cmake --build . --target test_unit -j 48 will do that.