e2e: Test breakage

[yanniszark]

/kind bug

What steps did you take and what happened:
E2E tests are not succeeding because they can't install the tf-operator manifests:

katib/test/scripts/v1beta1/setup-katib.sh

Line 70 in 1729997

cd "${MANIFESTS_DIR}/tf-training/tf-job-crds/base"

This is an artifact of using the manifests master branch to install the tf-operator.
A recent folder restructuring in manifests is what triggered it:
kubeflow/manifests#1738
kubeflow/manifests#1739

What did you expect to happen:
E2E tests should succeed.

Anything else you would like to add:
This is an artifact of using the manifests master branch to install the tf-operator.
As master is not guarranteed to be stable or avoid any folder renames, the e2e tests for katib broke.

These are the possible solutions:

• Pin the manifests version to the latest stable release (v1.2). This will prevent it from breaking in the future and I think is the best choice. We can update it regularly as with other deps.
• Change the directories they use to the new ones.

I recommend the first approach. If you agree, I can make a quick PR.

cc @kubeflow/wg-automl-leads

[andreyvelich]

I agree with the first approach since we can verify our e2e test on the latest stable training operator releases.
It would be great if you can submit the PR @yanniszark!

[andreyvelich]

Fixed by: #1438.