deploy(S4): Blender headless pipeline + WebAR client + backend fixes

backend/agent/react_agent.py

@@ -120,15 +120,16 @@ RULES:
     async def _llm_step_stream(self, prompt: str, context: dict) -> AsyncGenerator[ReActStep, None]:
         t0 = time.time()
         try:
-            provider = context.get("llm_provider", "OPENAI").upper()
-            api_key = context.get("llm_key", "")
-            
-            if not api_key and provider != "OLLAMA":
-                yield ReActStep("final_answer", f"Error: API Key missing for provider {provider}", duration_ms=int((time.time() - t0) * 1000))
-                return
-
+            # NOTE: the LLM key/provider are NOT taken from `context` here.
+            # stream_with_fallback() below resolves keys itself across the full
+            # provider chain (Gemini domain keys → Groq → OpenAI → Anthropic →
+            # Mistral → Ollama(PC) → NVIDIA vault) and always yields a response.
+            # A previous `context["llm_key"]` pre-check gated this path and
+            # bailed with "API Key missing for provider OPENAI" because
+            # agent_service never populates llm_key — dropping every desktop
+            # reply. Removed; the fallback chain is the single source of truth.
             system_prompt = self._build_system_prompt()
-            
+
             from backend.services.connectors import stream_with_fallback
             
             raw = ""