scaling parameter of Transformer always set to True

[StanislasGuinel]

The scaling parameter is always set to True when creating the TransformerTrainingNetwork or the TransformerPredictionNetwork but it is still a parameter of the TransformerEstimator.
Is there a reason why ?

[lostella]

Probably not. Without even looking into the details of how this option is used (and whether or not scaling=False makes sense), if the network exposes the option, and the estimator does too, it only makes sense to use whatever is set in the estimator.

Turning this into an issue, even though the fix should be trivial.