chore(llm): switch vllm model_candidates from Ouro to Phi-4-mini + Qwen2.5-3B

Ouro models incompatible with transformers 5.x bundled in cf env.
Phi-4-mini-instruct tried first (stronger benchmarks, 7.2GB);
Qwen2.5-3B-Instruct as VRAM-constrained fallback (5.8GB).

config/llm.yaml

@@ -28,7 +28,7 @@ backends:
     type: openai_compat
   ollama_research:
     api_key: ollama
-    base_url: http://host.docker.internal:11434/v1
+    base_url: http://ollama_research:11434/v1
     enabled: true
     model: llama3.1:8b
     supports_images: false
@@ -40,7 +40,7 @@ backends:
     type: vision_service
   vllm:
     api_key: ''
-    base_url: http://host.docker.internal:8000/v1
+    base_url: http://vllm:8000/v1
     enabled: true
     model: __auto__
     supports_images: false
@@ -48,12 +48,12 @@ backends:
     cf_orch:
       service: vllm
       model_candidates:
-        - Ouro-2.6B-Thinking
-        - Ouro-1.4B
+      - Phi-4-mini-instruct
+      - Qwen2.5-3B-Instruct
       ttl_s: 300
   vllm_research:
     api_key: ''
-    base_url: http://host.docker.internal:8000/v1
+    base_url: http://vllm:8000/v1
     enabled: true
     model: __auto__
     supports_images: false