feat(aml): ML-based false positive reduction (#394)

[bjabrack-29]

close #394

• Add logistic-regression scorer with SHAP-style feature attributions and human-readable justification for every suppression (audit-ready)
• Training pipeline consumes analyst TP/FP decisions via SGD on binary cross-entropy; persists versioned models to aml_model_versions
• Champion/challenger framework: shadow mode, deterministic A/B routing, auto-promotion when challenger achieves >=30% FP rate improvement
• PSI-based drift detection (critical at PSI>0.25) + accuracy degradation alerts; persisted to aml_drift_metrics
• MlAugmentedScreener wires ML layer into existing screening pipeline; sanctions hits are never suppressed regardless of model score
• Migration: aml_model_versions, aml_training_samples, aml_shadow_evaluations, aml_drift_metrics, aml_ml_scoring_audit
• Integration tests covering all 5 acceptance criteria

[drips-wave]

@bjabrack-29 Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits