Added TFServing JSON Support

[axsaucedo]

This pull request contains the code that adds support for raw JSON to be sent to the TF Server, so it allows for the more flexible interface that TF Serving provides. This issue addresses #694

[axsaucedo]

This has been tested locally and works as expected, but when run in the kubernetes cluster the TFServing image fails to pull from the Google bucket with the following error:

2019-07-16 18:22:10.905196: I tensorflow_serving/model_servers/main.cc:154] Building single TensorFlow model file config:  model_name: mnist-model model_base_path: gs://seldon-mnist-model-bucket/mnist-model
2019-07-16 18:22:10.905412: I tensorflow_serving/model_servers/server_core.cc:444] Adding/updating models.
2019-07-16 18:22:10.905441: I tensorflow_serving/model_servers/server_core.cc:499]  (Re-)adding model: mnist-model
2019-07-16 18:22:11.776606: I external/org_tensorflow/tensorflow/core/platform/cloud/retrying_utils.cc:77] The operation failed and will be automatically retried in 1.45458 seconds (attempt 1 out of 10), caused by: Unavailable: Error executing an HTTP request (error code 6, error message 'Couldn't resolve host name')
        Performing request. Detailed error: Couldn't resolve host 'metadata'

[axsaucedo]

@cliveseldon it seems that the container cannot reach the bucket. It's currently public, and it has been attempted with both the default bucket and a bucket created for this. Is there any pieces that I should be aware in regards to access for buckets? Are they only reachable from a GCloud cluster?

[axsaucedo]

The issue was the TFServing image which was outdated and causing issues. It now has been updated to tensorflow/serving:latest, and the changes have been reflected in the Notebook JSON + Helm. The example now provides support for JSON.

[ukclivecox]

predict_grpc_prev is an old method and should be removed

[axsaucedo]

This introduces the functionality required for #693