pub fn weighted_unscaled_predictions(unscaled: &[f64], weights: &[f64]) -> f64
_combine_predictions, NNLS-on path: each member’s prediction has already been inverse-transformed by the caller; combine via the fitted weights.
_combine_predictions