currently:

• we use cuda::std::complex throughout the code, when compiling with CUDA.
• we use std::complex throughout the code, when NOT compiling with CUDA.

Q: what if we used cuda::std::complex only in CUDA device code?

• all parts of the shader using complex would need to be template specialized for CPU and CUDA
• conversion between std::complex and cuda::std::complex would have to take place on transfers from/to device
• rayx-core library object would have the same API, regardless of CUDA enabled or disabled. probably makes rayx-python easier.

NOTE: there is more:

• std::optional vs cuda::std::optional
• std::variant vs cuda::std::variant
• ...
  https://nvidia.github.io/cccl/libcudacxx/standard_api/utility_library.html