Signature Selection: Multi-Objective Best Biomarker Selection

Functions for selecting the best biomarker signature from nested CV results. Implements multiple selection strategies: constrained 1SE rule, weighted scoring, and Pareto optimality for transparent multi-objective trade-offs.

Details

## Clinical Context

In biomarker discovery, selecting the "best" signature requires balancing: - **Performance** (AUC, accuracy): Predictive power - **Stability**: Reproducibility across resampling (Nogueira Index) - **Parsimony**: Fewer features = cheaper, more practical clinical panels

This module provides three selection modes: 1. `constrained_1se`: Conservative selection within 1 standard error of best 2. `weighted`: Explicit weighted combination of objectives 3. `pareto`: Return non-dominated solutions for human inspection