Pin MKL to 2020.04 (#2486)

Summary:
The version of MKL that is installed alongside PyTorch has been bumped
to 2022.1 on Windows and it is causing installation issue in unit tests.

This commit pins the previous version

Backport of https://github.com/pytorch/audio/issues/2463

Pull Request resolved: https://github.com/pytorch/audio/pull/2486

Reviewed By: nateanl

Differential Revision: D37160156

Pulled By: mthrok

fbshipit-source-id: 7e3a30c25782b349a3cad2ee6d1141affc921881

.circleci/unittest/windows/scripts/install.sh

@@ -29,7 +29,7 @@ else
     cudatoolkit="cudatoolkit=${version}"
 fi
 printf "Installing PyTorch with %s\n" "${cudatoolkit}"
-conda install -y -c "pytorch-${UPLOAD_CHANNEL}" -c conda-forge "pytorch-${UPLOAD_CHANNEL}"::pytorch[build="*${version}*"] "${cudatoolkit}" pytest
+conda install -y -c "pytorch-${UPLOAD_CHANNEL}" -c conda-forge "pytorch-${UPLOAD_CHANNEL}"::pytorch[build="*${version}*"] "${cudatoolkit}" 'mkl=2020.4' pytest
 
 torch_cuda=$(python -c "import torch; print(torch.cuda.is_available())")
 echo torch.cuda.is_available is $torch_cuda