fix(F-14b): add pickle/model integrity checking mechanism

chore: add latest training artifacts and data manifests from local retrain run

ci: trigger CI to verify all tests pass after LFS fix

fix: add clip_database.pkl and ensemble_results.json via LFS

fix: track model files via Git LFS, not gitignore

fix: remove model files from git tracking

feat: add trained model artefacts from initial pipeline run

feat(ensemble): — XGBoost trained classifier replaces hardcoded signal weights

docs: document manifest.csv is generated locally

data: add gitkeep for new real dataset folders

docs: add full dataset registry — 8 real + 10 AI sources, 200K target

data: scaffold dataset directory structure with gitignore rules

Fix: rebuild CLIP database with more balanced real/AI samples

Fix: rebuild CLIP database with balanced real/AI samples

Build CLIP reference database (#32)

Add reference dataset download scripts (#32)