Provenance-aware within-sample scoring methods

These functions implement scoring methods that explicitly account for cohort and specimen-provenance structure. All functions expect an expression matrix with samples in rows and features in columns. `meta` must have one row per sample in the same order as `expr`.

Details

The mixed-effects scorer is split into a training step and a prediction step: disease labels are used only by `fit_mixed_effects_scorer()` on the training pool. `predict_mixed_effects_scorer()` and `score_mixed_effects_scorer()` with a supplied `fit` object do not read disease labels from test metadata.