feat(p1x): de-hardcode VLM models/endpoints to vlm_config (DGX-ready)

Migre les call-sites VLM serveur vers la configuration centrale pour
fonctionner sur DGX (tunnel Ollama 11434), où gemma4:* est absent et le
port Docker 11435 est mort.

- task_planner, replay_verifier, domain_context, ir_builder, resolve_engine
  (popup): modele -> vlm_config.get_vlm_model(), defaut 11435 -> 11434
  (override GEMMA4_PORT legacy conserve)
- resolve_engine (grounding bbox x2): nouvel helper
  vlm_config.get_bbox_grounding_model() (var dediee RPA_BBOX_GROUNDING_MODEL,
  fallback RPA_GROUNDING_MODEL puis qwen2.5vl:7b-rpa) -> desambiguise le
  conflit D5-v3b, bbox_2d + num_ctx 4096 preserves
- safety_checks_provider: defaut -> get_vlm_model(), override
  RPA_SAFETY_CHECKS_LLM_MODEL preserve
- ui_detector: default_factory + resolution lazy (corrige aussi un gel a
  l'import), pas d'appel reseau a l'import
- field_extractor: property lazy via vlm_config

TDD strict (RED->GREEN), 305 tests verts, tests mockes HTTP (zero dependance
DGX reel), aucun alias Ollama.

Hors perimetre (arbitrage Dom): client Lea agent_v1/executor.py (gele),
chemin V4 observe_reason_act (RPA_REASONING_MODEL), core/config.py defaults.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/unit/test_ui_detector.py

@@ -2,17 +2,55 @@
 import pytest
 import sys
 from pathlib import Path
+from unittest.mock import MagicMock, patch
 from PIL import Image
 
 sys.path.insert(0, str(Path(__file__).parent.parent.parent))
-from core.detection.ui_detector import UIDetector
+from core.detection.ui_detector import UIDetector, DetectionConfig
 
 class TestUIDetector:
     def setup_method(self):
         self.detector = UIDetector()
-    
+
     def test_detector_initialization(self):
         assert self.detector is not None
 
+
+class TestVlmModelDehardcode:
+    """Le modèle VLM ne doit plus être codé en dur (gemma4:e4b) dans la config."""
+
+    def test_config_no_hardcoded_gemma4(self, monkeypatch):
+        """Sans env, DetectionConfig.vlm_model ne vaut plus le littéral gemma4:e4b."""
+        monkeypatch.delenv("RPA_VLM_MODEL", raising=False)
+        monkeypatch.delenv("VLM_MODEL", raising=False)
+        cfg = DetectionConfig()
+        assert cfg.vlm_model != "gemma4:e4b"
+        assert not cfg.vlm_model  # None ou "" → résolution déléguée à l'init
+
+    def test_config_respects_env(self, monkeypatch):
+        """RPA_VLM_MODEL est honoré dans la config."""
+        monkeypatch.setenv("RPA_VLM_MODEL", "mon-modele:test")
+        assert DetectionConfig().vlm_model == "mon-modele:test"
+
+    def test_initialize_vlm_resolves_lazily(self, monkeypatch):
+        """Sans modèle explicite, _initialize_vlm résout via vlm_config (pas de hardcode)."""
+        monkeypatch.delenv("RPA_VLM_MODEL", raising=False)
+        monkeypatch.delenv("VLM_MODEL", raising=False)
+
+        captured = {}
+
+        class FakeClient:
+            def __init__(self, endpoint=None, model=None):
+                captured["model"] = model
+
+        import core.detection.ui_detector as uidet
+        with patch.object(uidet, "check_ollama_available", return_value=True), \
+             patch.object(uidet, "OllamaClient", FakeClient), \
+             patch.object(uidet.vlm_config, "get_vlm_model", return_value="modele-resolu:test"):
+            UIDetector()
+
+        assert captured["model"] == "modele-resolu:test"
+
+
 if __name__ == '__main__':
     pytest.main([__file__, '-v'])