feat: support return raw prompt and input ids by parameters

Author: soaringk

rtp_llm/openai/renderers/deepseekv32_renderer.py

@@ -260,8 +260,15 @@ def _create_detector(
             DeepSeekV32Detector if tools are present, None otherwise
         """
         if request.tools:
-            # Pass the encoding module to use official parse_tool_calls
-            return DeepSeekV32Detector(encoding_module=self.encoding_module)
+            # Determine thinking_mode based on whether request is in thinking mode
+            thinking_mode = "thinking" if self.in_think_mode(request) else "chat"
+
+            # Pass the encoding module and thinking_mode to detector
+            # Detector is created fresh for each request (not singleton)
+            return DeepSeekV32Detector(
+                encoding_module=self.encoding_module,
+                thinking_mode=thinking_mode
+            )
         return None
 
     @override