Fix GPU continuous build: correct torchaudio version

torchaudio 0.17 doesn't exist, but 2.2 does.

Use a slightly older cuda version, since that makes cuda detection work on the machines we're running on (RTX 2080's on Ubuntu 18.04, which presumably aren't compatible with CUDA 12).

(I'm not sure if the tests actually need torchaudio or torchvision, since those aren't listed dependencies of any of our software, but let's at least install a valid one)

Passing yamato run with this change: https://unity-ci.cds.internal.unity3d.com/job/34869354/logs

.yamato/pytest-gpu.yml

@@ -11,10 +11,11 @@ pytest_gpu:
       python3 -m pip install pyyaml --index-url https://artifactory.prd.it.unity3d.com/artifactory/api/pypi/pypi/simple
       python3 -u -m ml-agents.tests.yamato.setup_venv
       python3 -m pip install --progress-bar=off -r test_requirements.txt --index-url https://artifactory.prd.it.unity3d.com/artifactory/api/pypi/pypi/simple
-      python3 -m pip install torch==2.2.1+cu121 torchvision==0.17.1+cu121 torchaudio==0.17.1 --index-url https://download.pytorch.org/whl/cu121
+      python3 -m pip install torch==2.2.1+cu118 torchvision==0.17.1+cu118 torchaudio==2.2.1+cu118 --index-url https://download.pytorch.org/whl/cu118
       if python -c "exec('import torch \nif not torch.cuda.is_available(): raise')" &> /dev/null; then
         echo 'all good'
       else
+        echo 'cuda device not available!'
         exit 1
       fi
       python3 -m pytest -m "not slow" --junitxml=junit/test-results.xml -p no:warnings