chore(dgx): snapshot consolidation WIP pour transfert poc DGX

Regroupe le WIP non committé requis pour le clone/runtime DGX (Option A) :
- api_stream.py : préflight replay + smoke santé modèles + handler 403 WP-B
- de-hardcode VLM : vlm_config, gpu/*, vram_orchestrator, ollama_manager
- stream_processor, semantic_matcher, agent_chat (app/planner/intent)
- workflows.db (acquis ; le transfert artifacts le mettra à jour + rewrite chemins)
- docs : plans DGX, benchmarks VLM/grounders, recherche SOTA, coordination 8 juin

Snapshot destiné à la branche poc-dgx poussée sur Gitea pour cloner le DGX.
Scan anti-secret : clean. graphify (repo embarqué) exclu.

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

tests/unit/test_replay_preflight.py

@@ -0,0 +1,78 @@
+"""Tests Job 1 — préflight replay non destructif.
+
+Vérifie les fonctions pures du préflight (`_detect_dialogs_static`,
+`_build_preflight_report`) et la garantie de non-destruction (aucune mutation des
+états replay). Le préflight prouve `commande → workflow → actions non vides →
+dialogues statiquement détectables` SANS injecter d'action ni poser de lock.
+"""
+
+import os
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+if str(ROOT) not in sys.path:
+    sys.path.insert(0, str(ROOT))
+
+os.environ.setdefault("RPA_AUTH_DISABLED", "true")
+
+from agent_v0.server_v1 import api_stream
+
+
+# --- Workflow dict minimal avec un node dialogue "Enregistrer sous" ---
+def _wf_with_save_dialog():
+    return {
+        "name": "Bloc-notes save test",
+        "nodes": [
+            {"node_id": "n1", "template": {"window": {"title_contains": "Bloc-notes"}}},
+            {"node_id": "n2", "template": {"window": {"title_contains": "Enregistrer sous"}}},
+        ],
+    }
+
+
+def _wf_without_dialog():
+    return {
+        "name": "Explorateur simple",
+        "nodes": [
+            {"node_id": "n1", "template": {"window": {"title_contains": "Explorateur"}}},
+        ],
+    }
+
+
+_ACTIONS = [
+    {"type": "key_combo", "keys": "ctrl+s"},
+    {"type": "text_input", "text": "test_lea.txt"},
+    {"type": "click", "x_pct": 0.45, "y_pct": 0.61},
+]
+
+
+def test_detect_dialogs_static_finds_save_as():
+    dialogs = api_stream._detect_dialogs_static(_wf_with_save_dialog())
+    assert any("enregistrer sous" in d.lower() for d in dialogs)
+
+
+def test_detect_dialogs_static_empty_when_none():
+    assert api_stream._detect_dialogs_static(_wf_without_dialog()) == []
+
+
+def test_build_preflight_report_shape():
+    report = api_stream._build_preflight_report(
+        _wf_with_save_dialog(), "wf_123", _ACTIONS
+    )
+    assert report["workflow_known"] is True
+    assert report["workflow_id"] == "wf_123"
+    assert report["n_actions"] == 3
+    assert report["action_types"]["key_combo"] == 1
+    assert report["action_types"]["click"] == 1
+    assert report["non_destructive"] is True
+    assert any("enregistrer sous" in d.lower() for d in report["dialogs_detected"])
+    # sample_actions limité (≤3) et présent
+    assert 0 < len(report["sample_actions"]) <= 3
+
+
+def test_build_preflight_report_does_not_mutate_replay_state():
+    before_q = dict(api_stream._replay_queues)
+    before_s = dict(api_stream._replay_states)
+    api_stream._build_preflight_report(_wf_with_save_dialog(), "wf_123", _ACTIONS)
+    assert dict(api_stream._replay_queues) == before_q
+    assert dict(api_stream._replay_states) == before_s

tests/unit/test_workflow_components.py

@@ -196,6 +196,22 @@ class TestSemanticMatcher:
         
         workflows = matcher.get_all_workflows()
         assert len(workflows) == 2
+
+    def test_load_workflows_recursively(self, temp_workflows_dir):
+        """Les workflows appris dans des sous-dossiers machine sont visibles."""
+        machine_dir = temp_workflows_dir / "DESKTOP-TEST_windows"
+        machine_dir.mkdir()
+        nested = {
+            "name": "Bloc-notes Enregistrer",
+            "description": "Workflow appris dans un sous-dossier machine",
+            "tags": ["bloc-notes"],
+        }
+        with open(machine_dir / "notepad_save.json", "w") as f:
+            json.dump(nested, f)
+
+        matcher = SemanticMatcher(str(temp_workflows_dir), use_llm=False)
+
+        assert matcher.get_workflow("notepad_save") is not None
     
     def test_find_workflow_exact_match(self, temp_workflows_dir):
         """Test matching exact."""
@@ -221,6 +237,52 @@ class TestSemanticMatcher:
         match = matcher.find_workflow("créer une facture")
         assert match is not None
         assert "Facturation" in match.workflow_name
+
+    def test_find_learned_save_workflow_from_node_text(self, temp_workflows_dir):
+        """Les textes appris dans les nodes alimentent le matching sémantique."""
+        workflow = {
+            "name": "Tâche Bloc-notes",
+            "description": "Auto-generated workflow",
+            "nodes": [
+                {
+                    "name": "State Pattern 0",
+                    "template": {
+                        "window": {
+                            "title_contains": "Sans titre – Bloc-notes",
+                        }
+                    },
+                },
+                {
+                    "name": "State Pattern 1",
+                    "template": {
+                        "window": {
+                            "title_contains": "Enregistrer sous",
+                        },
+                        "text": {
+                            "required_texts": ["Nom du fichier", "Enregistrer"],
+                        },
+                    },
+                },
+            ],
+            "edges": [
+                {
+                    "action": {
+                        "type": "click",
+                        "target_text": "Enregistrer",
+                    },
+                    "expected_window_title": "Enregistrer sous",
+                }
+            ],
+        }
+        with open(temp_workflows_dir / "notepad_save_as.json", "w") as f:
+            json.dump(workflow, f)
+
+        matcher = SemanticMatcher(str(temp_workflows_dir), use_llm=False)
+
+        match = matcher.find_workflow("sauvegarde le fichier notepad", min_confidence=0.2)
+        assert match is not None
+        assert match.workflow_id == "notepad_save_as"
+        assert "enregistrer" in matcher.get_workflow("notepad_save_as").keywords
     
     def test_extract_params(self, temp_workflows_dir):
         """Test extraction de paramètres."""