fix: LLM_MODEL default to openrouter/deepseek/deepseek-v4-flash

LiteLLM requires openrouter/ prefix to route through openrouter API.
Without it, deepseek/* models route to Deepseek native API and fail
with the openrouter API key.

.env.example

@@ -6,7 +6,7 @@ LOG_LEVEL=info
 # --- LLM (LiteLLM provider/model) ---
 LLM_PROVIDER=openrouter
 LLM_API_KEY=sk-or-v1-PLACEHOLDER-replace-before-deploy
-LLM_MODEL=anthropic/claude-3.5-sonnet
+LLM_MODEL=openrouter/deepseek/deepseek-v4-flash
 LLM_API_BASE=
 
 # --- MCP Tool Server ---