feat: use GGUF chat template via apply_chat_template

- Replaced manual prompt building with LlamaModel::apply_chat_template
- Uses model's baked-in Jinja template (system/user/assistant/tool format)
- Added <think> trigger after template for thinking mode
- Clean_text strips only <|im_end|>, <|im_start|>, <think>, </think>

src/presentation/handler/chat.rs

@@ -26,7 +26,8 @@ pub async fn chat_completions(
 ) -> Result<Response, AppError> {
     let max_tokens = req.max_tokens.unwrap_or(256).min(1024);
     let stop = req.stop.clone().unwrap_or_default();
-    let prompt = chat::build_prompt(&req.messages, &req.tools);
+    let prompt = chat::build_prompt(&state.engine.model, &req.messages, &req.tools)
+        .map_err(|e| AppError::LlmError(e))?;
 
     // Tokenize
     let input_tokens = state