Add options for GPU acceleration

[tcmetzger]

The Python user interface provides options for GPU-based acceleration. Specifically: if the Python interface detects a CuPy array as input, it would automatically call the respective OpenACC/Cuda kernels provided by https://github.com/earth-system-radiation/rte-rrtmgp-cuda-kernels). Supporting Cuda functions is an incremental process where we start by exposing a single Cuda function to start with, and then add more functions as time permits.