Implement PyTorch controller for multi node GPU scaling

[jaystary]

Use Case

We want to utilize multi node GPU scaling for PyTorch for a benchmark / potential larger scale model training

Ideas of Implementation

Implement KF Training Operator which as an added benefit should unlock all the relevant frameworks as well. https://github.com/kubeflow/training-operator

---

Message from the maintainers:

Excited about this feature? Give it a :thumbsup:. We factor engagement into prioritization.

[davidspek]

The PyTorch operator resource can already be used in the current Kubeflow deployment, so there's no need to pause development while we update the deployment to the training operator.