[BE] Use nested namespace in `ATen/cuda` (#106708)
<!--
copilot:poem
-->
### <samp>🤖 Generated by Copilot at ac9bd0c</samp>
> _We're sailing on the CUDA sea, with tensors and graphs aplenty_
> _We're refactoring the code, to make it clear and neat_
> _We're using nested namespaces, like `at::cuda::blas`_
> _So heave away, me hearties, heave away on the count of three_
Pull Request resolved: https://github.com/pytorch/pytorch/pull/106708
Approved by: https://github.com/kit1980, https://github.com/Skylion007
