Add CUDA backend support for the heterobatch interface

[YuWei-CH]

This PR adds native CUDA support for the heterobatch interface.
Specifically, it:

1. Translates the original CuPy-based implementation into native CUDA, implementing both the finufft_fast and finufft_chi2 interfaces directly in CUDA.
2. Integrates these CUDA kernels as helper implementations for the heterobatch interface.
3. Updates CMakeLists.txt to optionally enable the CUDA backend for the heterobatch interface when requested by the user.
4. Adds documentation for the CUDA backend to README.md, including build and configuration instructions.