Add install instructions for Windows GPU

[NicolasHug]

Thanks to @traversaro , Windows GPU support is available via conda-forge.

[NicolasHug]

Drive-by, I don't want to bother readers about installing FFmpeg from source, they shouldn't need to and it shouldn't be in the "entry point" install instructions anyway.

[traversaro]

Suggested change

	conda install torchcodec -c conda-forge
	conda install torchcodec -c conda-forge

this should indeed install torchcodec with cuda support on a machine with cuda support, but it may silently install the cpu version on a machine without cuda support. If you want to make sure that the cuda version is installed, and an error is printed in case it is not possible to install, you need to:

Suggested change

	conda install torchcodec -c conda-forge
	conda install torchcodec=*=*cuda* -c conda-forge

[traversaro]

As a meta-comment on this, I do not monitor all the issues in the torchcodec repo, so if there are users experiencing problems with the conda-forge packages, feel free to ping me or to redirect them to open an issue at https://github.com/conda-forge/torchcodec-feedstock/issues .

[traversaro]

Also, I am not sure if the instructions in the current form actually works. In point 1, it is described how to install PyTorch with pip, but that would not work out of the box (due to different ABIs) with the conda-forge torchcodec package. If one wants to use CUDA-enabled torchcodec on Windows, it can simply create a new conda environment with:

conda create -n torchcodeccuda -c conda-forge torchcodec=*=*cuda*

and this command will automatically install the approprite cuda, ffmpeg and pytorch version via conda-forge.

[scotts]

@traversaro, thank you for making this happen!

[traversaro]

@traversaro, thank you for making this happen!

Thanks for for all the work on torchcodec, I actually did not changed anything for the test to pass on Windows+CUDA, everything worked out of the box on the first try.