Score Transform for filter ensemble

[mb706]

We could have a "score transform" function hyperparameter for the ensemble

so currently ensemble_score := sum_{filter i} scores_by_i * weight_i

and then we could do ensemble_score := sum_{filter i} score_transform(scores_by_i) * weight_i

inspired by https://github.com/ck37/featurerank/blob/56e1426ffac84d0512c9e7a502be535f6ea57f09/R/aggregators.R#L2 and @bblodfon

[mb706]

score_transform would need to vectorize over score-vector produced by filter for all p task features and would need to output a vector of same length

[mb706]

Actually need two functions here: filter_score_transform (initialized to identity, applied before weighted mean, but after rank transform), and result_score_transform (applied to output).