Aplr

Now it is possible to optionally provide the following for each predictor:
- Learning rate.
- Penalty for non-linearity.
- Penalty for interactions.

- Speeded up the algorithm when penalty_for_non_linearity=1.0 or penalty_for_interactions=1.0.
- Automatically rounding penalty_for_non_linearity and penalty_for_interactions to the nearest boundary in the [0.0, 1.0] range if the user specifies something outside of this range.

Added two constructor parameters to help controlling the interpretability versus predictiveness trade-off:
- penalty_for_non_linearity (default = 0.0). Specifies a penalty in the range [0.0, 1.0] on terms that are not linear effects. A higher value increases model interpretability but can hurt predictiveness.
- penalty_for_interactions (default = 0.0). Specifies a penalty in the range [0.0, 1.0] on interaction terms. A higher value increases model interpretability but can hurt predictiveness.

Added the possibility to use linear effects only for a custom number of initial boosting steps. This can be used for example to increase interpretability by building models that place more weight on linear effects.

Fixed a minor bug that sometimes added unnecessary terms to the model.

Fixed a minor (and for practical purposes insignificant) inconsistency w.r.t. rounding errors.