New repo xlnscuda provides CUDA support

[markgarnold]

I just translated a subset of xlns16.cpp and xlns32.cpp to support identical operations on a CUDA __device__. These are in the new repository xlnscuda. Although this is not needed for the current GSoC LLM ggml project, it suggests if an LNS-CPU backend is created for ggml, a similar one can be created for LNS-CUDA without much additional effort.