[ML] Move computation of loss value in a single place

* Right now `captureBestHyperparameters` has a side-effect of defining a business logic of how the loss function needs to be adjusted wrt. model size penalty.
* For incremental learning, we add another 0.01 times the difference in the old predictions
* We add other model size penalties at multiple other places.

This leads to the situation that we have to follow the execution path to find out how the loss function value is formed. Since it is a critical part of the ML algorithm, I think this business logic should be defined in one place.

Start by tracing captureBestHyperparameters and the incremental-learning path to locate every adjustment to the loss value, including model-size penalties and the old-prediction term. Define a single loss-computation location that preserves the existing adjustments, then verify that all callers use it and that the ML tests cover the resulting value.

Scritto dal modello di indicizzazione a partire dal testo della issue.