This PR always enables CUDA language support in pylibraft's CMakeLists.txt, to work around an issue I encountered while building.

I am trying to build pylibraft in rapids-compose and saw an error in the CMake support for enabling CUDA.

CMake Error at ~/rapids1/compose/etc/conda/cuda_12.0/envs/rapids/share/cmake-3.26/Modules/CMakeDetermineCUDACompiler.cmake:279 (message):
  CMAKE_CUDA_ARCHITECTURES:

    NATIVE

  is not one of the following:

    * a semicolon-separated list of integers, each optionally
      followed by '-real' or '-virtual'
    * a special value: all, all-major, native

Call Stack (most recent call first):
  ~/rapids1/raft/cpp/build/release/raft-config.cmake:171 (enable_language)
  CMakeLists.txt:39 (find_package)

This error indicates that rapids_cuda_init_architectures(pylibraft) isn't being called, which handles some preprocessing for this architectures variable so that CMake recognizes the results.

It seems like the error shown below comes from this part of libraft's CMakeLists.txt where the CUDA language is being enabled without having first called rapids_cuda_init_architectures.

Currently, rapids_cuda_init_architectures is called and the CUDA language is only enabled in pylibraft when raft_FOUND is false. However, I'm building in an environment where raft can be found. It seems like we always need to enable CUDA support due to a requirement in libraft itself, based on the code I see in cuML and cuGraph. Therefore, I copied a similar change into pylibraft in this PR, and it appears to work.