Add user instruction to adjust cuda-version dependency during install if not running cuda 13

[benf549]

The current installation suggestion (conda install -c conda-forge nvmolkit) will fail to install with a gpu-accelerated PyTorch version (the cpu only version is installed) if the user is not running CUDA 13 (at least on my machine). After some trial and error, I was able to get a cuda accelerated PyTorch to install on my machines (running CUDA 12.4 and CUDA 12.9) by running the following commands:

CUDA 12.4 machine (12.4 seems to work w/ 12.6 which is the minimum version supported by nvmolkit according to conda):

conda install python=3.11 nvmolkit cuda-version=12.6

CUDA 12.9 machine:

conda install python=3.11 nvmolkit cuda-version=12.9

It would be helpful if there was an indication in the README that cuda-version may need to be adjusted to match what the user has installed (checkable with nvcc --version) or if this was somehow handled automatically. Running conda install -c conda-forge nvmolkit was installing cuda-version=13 which was not installed on either of my systems.

[evasnow1992]

Thank you for reporting this! We will update the README with clearer instructions on specifying cuda-version during installation. We will also update the conda-forge recipe so that the solver automatically selects the correct CUDA version to match your system. This fix will be included in the next release.