Bug: cusolver.py fails to find library in wsl2 miniconda3/conda environments using nvidia-cusolver-cu12

It actually isn't an issue, but a concern I want to raise for the devs of GPU4Pyscf.

The current implementation uses ctypes.util.find_library('cusolver'). In many modern Python environments (like Conda on WSL2), libraries installed via pip install nvidia-cusolver-cu12 are located inside the site-packages/nvidia/ subdirectories. find_library does not scan these paths, leading to an ImportError or AttributeError regarding missing symbols (e.g., cusolverDnDsygvd_bufferSize).

This leads to "lubcusolver.so.11 not found error " even when it is explicitly installed in isolated modern wsl2/conda environments where the packages are in a different subfolder system.

This legacy find_library works perfectly fine for non-isolated systems, but fails in modern isolated containers/conda environments adn others....

Suggested Fix for the devs: Instead of find_library, use a relative path lookup from the nvidia.cusolver module or use importlib.resources to locate the .so file within the package directory.

Start by inspecting cusolver.py and reproducing the import in a WSL2 Conda environment with nvidia-cusolver-cu12 installed. Check how the current library lookup behaves when the shared library is under the nvidia package directories. Done means the library is found and symbols such as cusolverDnDsygvd_bufferSize load without an ImportError or AttributeError.