Option to scale loss by `n`

[tlienart]

In case users prefer this;

that means though that things should be ScaledLosss{...} in fit methods etc, it shouldn't change much otherwise.