[skip ci] Add explanation for why we split TORCH_CUDA_API (#55641)
Summary:
Provide explanation for why we have (and use) the BUILD_SPLIT_CUDA option as a result of PR https://github.com/pytorch/pytorch/pull/49050.
This should hopefully clarify why there is both TORCH_CUDA_CU_API and TORCH_CUDA_CPP_API.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55641
Reviewed By: samestep
Differential Revision: D27661729
Pulled By: janeyx99
fbshipit-source-id: a68b44df2b45ce10590b9b0229558a1ad40ce485
