PyTorch Operator: Move manifests development upstream

[yanniszark]

Background

Umbrella-Issue: kubeflow/manifests#1740

As part of the work of wg-manifests for 1.3
(kubeflow/manifests#1735), we are moving manifests
development to upstream repos. This gives the application developers full
ownership of their manifests, tracked in a single place.

To give an example from the notebook-controller, the diagram shows the current
situation. Manifests are in two repos, blurring separation of responsibilities
and making things harder for application developers.

Instead, we will copy all manifests from the manifests repo back to each
upstream's repo. From there, they will be imported in the manifests repo. The
following diagram presents the desired state:

Current State

PyTorch Operator has manifests:

• In the manifests repo, under folder apps/pytorch-job/upstream.
• In upstream repo (https://github.com/kubeflow/pytorch-operator) under folder manifests.

Desired State

The proposed folder in the upstream repo to receive the manifests is:
manifests.

The goal is to consolidate all manifests development in the upstream repo.
The manifests repo will include a copy of the manifests under apps/pytorch-job/upstream.
This copy will be synced periodically.

Success Criteria

The manifests in apps/pytorch-job/upstream should be a copy of the upstream manifests
folder. To do that, the application manifests in kubeflow/manifests should be
moved to the upstream application repo.

/assign @yanniszark
cc @kubeflow/wg-training-leads