Sanity check APIs from Microsoft.ML.Tensorflow nuget

[codemzs]

We plan to GA Microsoft.ML.Tensorflow nuget in 1.2 release. Lets do a final pass on the binary and make sure we are only exposing relevant API.

[eerhardt]

Looking myself, I see the following places that require attention. There may be more, but these popped out to me.

1. TensorFlowUtils is a public class, but it doesn't have any externally visible members:

machinelearning/src/Microsoft.ML.TensorFlow/TensorFlow/TensorflowUtils.cs

Line 17 in f0f34ac

public static class TensorFlowUtils

We should make it internal.

2. TensorFlowModel contains the most important method of the component - ScoreTensorFlowModel. However, it also contains a RetrainTensorFlowModel whose documentation say it is "experimental":

machinelearning/src/Microsoft.ML.TensorFlow/TensorFlowModel.cs

Lines 105 to 108 in f0f34ac

	/// <remarks>
	/// The support for retraining is experimental.
	/// </remarks>
	public TensorFlowEstimator RetrainTensorFlowModel(

I think it is confusing to have an "experimental" API on the same class as a stable API.

[codemzs]

Thanks @eerhardt, below is the APIsurface after hiding unnecessary stuff. Retrain can wait, we are planning to expose more advance TF scenarios and retrain will come there with TF.NET backend.