Add interfaces for MLModel, MLModel.ModelInput, MLModel.ModelOutput generated classes

**Is your feature request related to a problem? Please describe.**
I have multiple models trained on a dataset that have multiple labels. I disable all of them and enable one of them at a time, do the training, and then enable only the next one, do the training etc.
I end up with multiple trained models and their respective generated classes for consumption.

**Describe the solution you'd like**
It would be great to have a common interface for the MLModel, MLModel.ModelInput and MLModel.ModelOutput classes, because when I need to consume them that would simplify my code significantly.

**Describe alternatives you've considered**
I can still use them separately, but this causes code duplication and code management difficulties.

**Additional context**
I would only need a definition on the interface of the MLModel for the `IModelOutput Predict(IModelInput input)` method, and the `Score` property on the interface of the `ModelOutput`.

Beginnen Sie damit, zu ermitteln, wie die generierten MLModel-, MLModel.ModelInput- und MLModel.ModelOutput-Klassen erzeugt werden, und untersuchen Sie anschließend die zugehörigen Einstiegspunkte der Generierung. Die Arbeit ist abgeschlossen, wenn die generierten Modelle die angeforderten gemeinsamen Schnittstellen bereitstellen, einschließlich Predict(IModelInput input) und der ModelOutput-Eigenschaft Score, ohne verbraucherspezifischen Code zu duplizieren.

Vom Indexierungsmodell aus dem Issue-Text verfasst.