v0.2.0

Added

• NoOpCorrector: identity baseline corrector for use in AutoCorrector / benchmark candidate lists.
• AutoCorrector: sklearn-compatible meta-corrector with method as a tunable hyperparameter, so GridSearchCV can sweep across corrector families. Accepts METHOD_REGISTRY aliases or a BaseBatchCorrector subclass; forwards only kwargs the inner method supports; proxies fitted attributes.
• BatchCorrectionBenchmark: configure-then-.fit diagnostic comparison with full_data / stratified_split protocols and optional bootstrap CIs (resample_metric or refit). Exposes results_, results_long_, corrected_, baseline_, .rank(), .to_dataframe(), .plot(); silhouette_batch ranked by |value|.
• maldibatchkit.metrics: per-batch sklearn metrics aggregated with weights='uniform' | 'balanced' | 'size' (or custom). Ships batch_{roc_auc, average_precision, balanced_accuracy, matthews_corrcoef, f1, precision, recall}_score; drops degenerate single-class folds with a warning.
• make_batch_scorer(batch, metric=..., weights=...): factory yielding a GridSearchCV-compatible scorer that slices batch per fold; supports all response_method values and string/callable metrics.