XLA support for CPU depends on CUDA

[matbme]

In src/ale/python/CMakeLists.txt, XLA support for both CPU and GPU depend on CUDA being installed. However, src/ale/python/ale_vector_xla_interface.cpp already checks whether __CUDACC__ is defined before creating bingings for GPU, so disabling XLA support entirely seems unnecessary in this case. Is there a particular reason for this?

[pseudo-rnd-thoughts]

Thanks for spotting that @matbme, I think we should relax the src/ale/python/CMakeLists.txt to be able to build XLA if BUILD_VECTOR_XLA_LIB is included in the build args.
@matbme Would you be able to make a PR to fix this?