Fix: Try direct deepseek model without openai prefix - let llm_params handle HF router format

configs/frontend_agent_config.json

@@ -1,5 +1,5 @@
 {
-  "model_name": "openai/google/gemini-2.5-flash",
+  "model_name": "deepseek/deepseek-chat-v3-0324",
   "save_sessions": true,
   "personal_trace_repo_template": "{hf_user}/ml-intern-sessions",
   "share_traces": true,