v0.2.2

DatasetDriftSummary with per-column signal attribution

Added

• DatasetDriftSummary — dataset-level aggregate with per-metric column attribution (jaccard_worst_column, null_rate_worst_column, etc.)
• Schema drift (missing_columns, new_columns) and row_count_delta computed in base interpreter from column profiles
• DatasetDriftSummary exported from top-level package

Changed

• InterpretationResult.drift_signals renamed to dataset_drift_summary
• aggregate_signals signature: list[DriftSignals] → dict[str, DriftSignals]
• compute_profile_signals returns dict[str, DriftSignals] (per-column only)
• to_dict/from_dict round-trip all DatasetDriftSummary fields