Variant-confidence v0.1.0 adds calibration layer for pathogenicity scores

Variant-confidence wraps existing variant-effect predictors like AlphaMissense to produce calibrated, uncertainty-aware outputs. It uses Platt scaling or isotonic regression, conformal prediction, and reports Expected Calibration Error with confidence intervals. The tool is open-source under AGPL-3.0 and has passed 28 tests in a clean audit.