Update Dockerfile to enable CUDA pytorch

[Fmstrat]

With the current Docker build, conda defaults to installing pytorch in CPU mode despite the use of pytorch and nvidia channels. This may be a change in how conda handles package relationships. The below command adjustment passes in the specific version numbers, which ensures that conda recognizes the use of pytorch-cuda as related, and properly installs the CUDA version of pytorch.

This can be validated after-the-fact with starting the container and running:

cd /app
conda activate tortoise
python -c "import torch; print(torch.cuda.is_available());torch.zeros(1).cuda()"

The current image returns an error, while after this PR, returns "True".

Note: This also updates CUDA in pytorch to 12.1 to better align to the CUDA docker image being used as the base.