refactor: réorganisation référentiels, nouveaux modules extraction, nettoyage code obsolète

- Réorganisation data/referentiels/ : pdfs/, dicts/, user/ (structure unifiée)
- Fix badges "Source absente" sur page admin référentiels
- Ré-indexation COCOA 2025 (555 → 1451 chunks, couverture 94%)
- Fix VRAM OOM : embeddings forcés CPU via T2A_EMBED_CPU
- Nouveaux modules : document_router, docx_extractor, image_extractor, ocr_engine
- Module complétude (quality/completude.py + config YAML)
- Template DIM (synthèse dimensionnelle)
- Gunicorn config + systemd service t2a-viewer
- Suppression t2a_install_rag_cleanup/ (copie obsolète)
- Suppression scripts/ et scripts_t2a_v2/ (anciens benchmarks)
- Suppression 81 fichiers _doc.txt de test
- Cache Ollama : TTL configurable, corrections loader YAML
- Dashboard : améliorations templates (base, index, detail, cpam, validation)

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

tests/test_splitter.py

@@ -192,7 +192,7 @@ class TestSplitDocuments:
 # --- Test intégration process_pdf ---
 
 class TestProcessPdfMulti:
-    @patch("src.main.extract_text_with_pages")
+    @patch("src.main.extract_document_with_pages")
     @patch("src.main.extract_medical_info")
     @patch("src.main._run_edsnlp", return_value=None)
     @patch("src.main._use_edsnlp", False)
@@ -203,9 +203,14 @@ class TestProcessPdfMulti:
         from src.main import process_pdf
         from src.config import DossierMedical, Diagnostic
         from src.extraction.page_tracker import PageTracker
+        from src.extraction.pdf_extractor import ExtractionStats
 
-        # Mock extract_text_with_pages retournant un texte multi-épisodes Trackare
-        mock_extract.return_value = (TRACKARE_MULTI, PageTracker([(0, len(TRACKARE_MULTI))]))
+        # Mock extract_document_with_pages retournant un texte multi-épisodes Trackare
+        mock_extract.return_value = (
+            TRACKARE_MULTI,
+            PageTracker([(0, len(TRACKARE_MULTI))]),
+            ExtractionStats(total_pages=1, chars_per_page=[len(TRACKARE_MULTI)], total_chars=len(TRACKARE_MULTI)),
+        )
 
         # Mock extract_medical_info retournant un DossierMedical minimal
         mock_medical.return_value = DossierMedical(