Updating the cloudml-template to TensorFlow v1.13

[ksalama]

This is a large, yet important PR, to update the cloudml-template, along with its associated examples, to use TensorFlow v1.13 APIs. The update includes:

• Simplifying the metadata.py
• Use the TF 1.13 APIs in the canned estimators
• Remove the use of custom estimators
• Renaming input.py to inputs.py, since it was causing issues when importing
• Creating a separate experiment.py module, and leave task.py module as the entry point only.
• Exporting an eval_saved_model to be used by TFMA
• Using an improved input_fn implementation
• Using MirroringStrategy if multiple GPUs found
• Eliminating the use of tf.contrib
• Restructuring the examples directory
• Updating README.md

---

This change is