Installing PyTorch and fastai as dependencies takes a lot of time and bandwidth

[sctse999]

Since these depdencies aren't absolutely necessary, should we separate them into another yml.

cd donkeycar
conda env create -f install\envs\windows.yml

In case anyone want to use it, they could do a separate installation like this:

conda env update --name donkey --file pytorch.yml --prune

This will help users not using pytorch much faster to complete the standard installation.

Reference: https://stackoverflow.com/questions/42352841/how-to-update-an-existing-conda-environment-with-a-yml-file

[sctse999]

https://github.com/autorope/donkeycar/pull/1066 submitted. Please review.