Forces CUDA Device 0 about libmolgrid

[dkoes]

With 0.5.1 more informative error messages are generated if a torch tensor isn't on the current GPU device and helper functions (molgrid.set_gpu_device and molgrid.get_gpu_device) are now available for changing the current GPU.

I think it is best practice to make the programmer explicitly decide what GPU things are on, so calling molgrid.set_gpu_device is going to be the solution rather than implicitly changing GPU devices to match the tensor.

[RMeli]

Are molgrid.set_gpu_device and molgrid.get_gpu_device documented anywhere? I think it would be beneficial to add molgrid.set_gpu_device to the tutorial (and the documentation if not already there, I was unable to find them).

[dkoes]

I needed to add these methods to the sphinx rst file for them to show up in the documentation. I've pushed the change. I'd gladly accept a pull request for a more comprehensive getting started tutorial (it's on the todo list, but happy for someone else to do it)...

[RMeli]

Thanks, it will be very useful to have these functions documented.

I almost finished a PyTorch-based implementation of the GNINA scoring function (to be extended for flexible docking) so maybe a subset of that could be used as a starting point for a more comprehensive tutorial? (However I used PyTorch-Ignite quite heavily, which might obfuscate molgrid concepts...)

I've been wanting to put my own notes about GNINA and molgrid into some sort of "user guide" but I haven't yet had time to think more concretely about it either.