Universal Operator Phase II: Reduce duplication and extract common utilities

[Jeffwan]

In Phase I #1298, we try to replicate codes from different operator directly and make them work without further optimization. This is part of kubeflow/common#138

However, this is not the ultimate goal. We can see there're lots of duplications and inefficient usage. in Phase II, we want to

1. Eliminate difference between operators and make them have same coding structure (including apis, docs, openapi specs)
2. Extract common methods to utilities which can be shared by all operators.
3. Simplify controller code structures. Merge pods.go service.go job.go into one file which is the core JobController implementation.
4. Make all operator pass tests and manually run tests against new operator

[Jeffwan]

The last part of this PR is

1. Fix all the issues to run multiple controllers correctly. Controller can not successfully run due to conflicts index fields #1312
2. Run more testing to make sure current behavior is same as original operator

[Jeffwan]

We need to fix these critical bugs to make sure old tensorflow controller work as well.

#1312
#1314
#1316
#1317

This make us comfortable to merge this branch to master and also make both old & new operator works together in one code base

[Jeffwan]

We can close this issue now.

/close

[google-oss-robot]

@Jeffwan: Closing this issue.

In response to this:

We can close this issue now.

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.