feat(phase2): Multi-signal NER — BDPM gazetteers, confiance EDS, safe patterns, GLiNER

Chantier 1: Intégration BDPM (5737 médicaments officiels) dans medication whitelist
Chantier 2: Safe patterns contextuels (dosages mg/mL/cpr, formes pharma, même ligne)
Chantier 3: Scores de confiance NER réels (edsnlp 0.20 ner_confidence_score)
Chantier 4: GLiNER zero-shot (urchade/gliner_multi_pii-v1) en vote croisé
Chantier 5: Scripts export silver annotations + fine-tuning CamemBERT-bio

0 fuite, 0 régression, -18 FP supplémentaires éliminés.
Sécurité: GLiNER ne peut rejeter que si confiance NER < 0.70.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

eds_pseudo_manager.py

@@ -64,6 +64,12 @@ class EdsPseudoManager:
             self._nlp = edsnlp.load(path)
         else:
             self._nlp = edsnlp.load(model_id_or_path)
+        # Activer les scores de confiance NER (edsnlp >= 0.16)
+        try:
+            ner_pipe = self._nlp.get_pipe('ner')
+            ner_pipe.compute_confidence_score = True
+        except Exception:
+            pass  # versions plus anciennes sans support confiance
         self._loaded = True
 
     def unload(self) -> None:
@@ -100,12 +106,15 @@ class EdsPseudoManager:
                 mapped = EDS_LABEL_MAP.get(label, None)
                 if mapped is None:
                     continue
+                # Score de confiance réel si disponible (edsnlp >= 0.16)
+                raw_score = getattr(ent._, 'ner_confidence_score', None)
+                conf = raw_score if isinstance(raw_score, float) else 1.0
                 ents.append({
                     "entity_group": label,
                     "word": ent.text,
                     "start": ent.start_char,
                     "end": ent.end_char,
-                    "score": 1.0,  # edsnlp ne fournit pas de score de confiance
+                    "score": conf,
                     "eds_mapped_key": mapped,
                 })
             out.append(ents)