Add a simple CUDA test to the CI

[cary-ilm]

Add a workflow run that builds a simple CUDA program that uses Imath, to validate that it compiles and runs.

[XueSongTap]

Hi! I'd be happy to help implement this issue and set up a CUDA-enabled CI workflow for Imath.

However, I'm currently not sure about the best way to implement it, since GitHub-hosted runners do not have access to NVIDIA GPUs.

That said, some projects use workarounds like compiling inside nvidia/cuda Docker containers (without runtime support), or use external runners with GPUs.

Let me know your thoughts — I'd love to contribute!

[cary-ilm]

Thank you for the interest!

Even just validating that it compiles properly would be sufficient, so just skip the execution altogether. We've had regressions where we've broken the nvcc build, so a test to catch that would be helpful.

We've also had errors in the HOST_DEVICE declaration that prevent certain symbols from being visible in CUDA, so a test program that includes the Imath headers and invokes as many functions as possible would catch those. There may actually be outstanding omissions, see #370.