Improve dynamic CPU/GPU choice in FFT evaluation and interpolation

[MegaRedHand]

We could maybe use a trait or enum to replace IsFFTField::field_name. The idea is to prioritize running on GPU, defaulting to CPU execution if it's not possible.

[peter941221]

Hi @MegaRedHand — I have access to a local RTX 5090 CUDA setup and this looks like a nice candidate for a small first contribution.

Rather than trying to redesign the abstraction all at once, would a first PR scoped to a minimal GPU-if-available / CPU-fallback path for FFT evaluation and interpolation be the most useful cut here?

If that direction makes sense, I’d be happy to take that smaller slice first and validate the GPU path on real hardware locally.