feat(training): add retrain-on-drift CLI

feat(domain): add retrain-on-drift gate (pure decision)

feat(mlflow): swap default tracking URI to sqlite:///mlflow.db

feat(api): make MLflow registry the serving source-of-truth

feat(adapters): add MLflowRegistryAdapter + load_from_registry_or_fallback

feat(api): surface registry version in /health

feat(training): register AraBERT LoRA in MLflow Model Registry

feat(training): register CatBoost baseline in MLflow Model Registry

feat(training): add polymorphic SentimentPyfunc + register_as in log_run

chore(phase): close Phase 5 — dialect breakdown (Gulf vs MSA F1, lexicon-v1)

feat(training): include dialect_breakdown in AraBERT LoRA report

feat(training): include dialect_breakdown in CatBoost baseline report

feat(training): add compute_dialect_breakdown utility

feat(domain): add Dialect enum + lexicon-based classify_dialect

feat(training): backfill confidence_histogram into reports JSON

feat(api): wire /predict drift recording + /metrics/drift endpoint

feat(adapters): add InMemoryDriftMonitor with predicted-class + confidence buckets

feat(domain): add DriftMonitorPort + PSI implementation + confidence bucketing

feat(api): rewire /predict through Depends(get_classifier) with hardened error handling

refactor(api): introduce create_app() + lifespan with SENTIMENT_BACKEND dispatch

chore(phase): close Phase 2 — AraBERT LoRA beats CatBoost baseline

style: apply ruff format to phase-02 adapter + training + tests

feat(demo): add --backend flag to showcase LoRA adapter alongside CatBoost

feat(training): add AraBERT LoRA training pipeline with MLflow tracking

feat(training): add mlflow_logging helper for file-backed run tracking

feat(adapters): add AraBERT LoRA adapter implementing SentimentClassifierPort

chore(phase): close Phase 1 — CatBoost baseline meets F1 bar

feat(demo): add CLI showcase for trained CatBoost adapter

feat(training): add CatBoost baseline training pipeline with F1+confusion matrix

feat(adapters): add CatBoost+TF-IDF adapter implementing SentimentClassifierPort

feat(adapters): add HARD dataset loader with 5★→3-class mapping

chore(repo): bootstrap walking skeleton and governance