Revamp the container build procedures

• Upgrade Drivers to 535 or 545. (nodes seem to work pretty stably)
• Upgrade NVIDIA Container Toolkit across all nodes. Then, update k8s-device-plugin.
• Move GitLab Runners to non-NVIDIA nodes
• Base foundation container to Ubuntu, not the NVIDIA CUDA base image.
• Install PyTorch with pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 and TensorFlow with pip3 install tensorflow[and-cuda]. Keras doesn't need to be installed separately and TensorRT installation is incorrect. Will be fixed through tensorflow[and-cuda].
• Add JAX, requires CUDA > 11.8.
• Consider best practices for buildx. Consult: https://github.com/docker/build-push-action
• Install GitHub Copilot and Copilot Chat manually by downloading and running code-server --install-extension <path to vsix>.
• Find a way to automatically add Jupyter server credentials to VS Code as well (allows running Jupyter notebooks in VS Code).
• Add Chrome and a virtual desktop (Selkies) to containers.