Refactor NAS RL Suggestion

[andreyvelich]

I Refactored NAS RL Suggestion code. I used https://github.com/google-research/google-research/tree/master/enas_lm in additional to original ENAS repository, since we will implement RNN search for Penn Treebank dataset.

Made few changes:

1. Changed algorithm settings to be more clear to user and deleted them from original yaml files. We have to add documentation about these algorithm settings, for example in ENAS README.
2. Deleted Controller lstm_num_layers. We were using only 1 layer of controller lstm, so it was useless. In the future, if we decide to implement this functionality, we can extend controller.
3. Removed Trainer and LSTM file. Left only useful functions in Controller.
4. Integrated Controller training. I think we should Train Controller after each succeeded Trial. User can specify controller_train_steps for how many steps Controller must be trained after each getSuggestion call.

5. Modified unit test with relevant example.

In the future PRs I:

1. Change NAS RL to ENAS, we have issue for that.
2. Modify UI with new Algorithm Settings.
3. Check Algorithm Settings in ValidateAlgorithmSettings function.
4. Enable to add None Algorithm Settings.

/assign @johnugeorge @gaocegege

[kubeflow-bot]

This change is 

[johnugeorge]

/lgtm
/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: johnugeorge

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [johnugeorge]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment