Need a 'workout test' for estimators

[Zruty0]

Pretty much any IEstimator needs to adhere to certain properties, which we should enforce in a test:

Estimator est:

• est.Fit() throws on a data if est.GetOutputSchema() throws on its schema.
  • The opposite may not be true.
• est.Fit() on the same data produces identical transformers.
• Schema validation errors should look properly (same message).

Fitted transformer xf = est.Fit(data):

• xf should be able to transform the data it was fitted on.
• xf.GetOutputSchema() should match est.GetOutputSchema().
• xf.Transform() throws on a data if xf.GetOutputSchema() throws on its schema.
  • The opposite may not be true.
• Save/load should be transparent for xf.
• Subject the outputs of xf.Transform() to the same workouts as we do for transforms: CheckSameSchema, CheckSameValues.

[Zruty0]

Actually, we also want to port TestDataPipe tests to ML.NET.