refactor(live): Use send_tool_response for function responses

The Gemini session now has a dedicated `send_tool_response` method, which is used instead of wrapping function responses in a `LiveClientToolResponse` and calling `send`.

`send` is deprecated.

Co-authored-by: Liang Wu <wuliang@google.com>
PiperOrigin-RevId: 899796634

Author: wuliang229

src/google/adk/models/gemini_llm_connection.py

@@ -105,10 +105,8 @@ async def send_content(self, content: types.Content):
       # All parts have to be function responses.
       function_responses = [part.function_response for part in content.parts]
       logger.debug('Sending LLM function response: %s', function_responses)
-      await self._gemini_session.send(
-          input=types.LiveClientToolResponse(
-              function_responses=function_responses
-          ),
+      await self._gemini_session.send_tool_response(
+          function_responses=function_responses
       )
     else:
       logger.debug('Sending LLM new content %s', content)

tests/unittests/models/test_gemini_llm_connection.py

@@ -118,10 +118,10 @@ async def test_send_content_function_response(
 
   await gemini_connection.send_content(content)
 
-  mock_gemini_session.send.assert_called_once()
-  call_args = mock_gemini_session.send.call_args[1]
-  assert 'input' in call_args
-  assert call_args['input'].function_responses == [function_response]
+  mock_gemini_session.send_tool_response.assert_called_once()
+  call_args = mock_gemini_session.send_tool_response.call_args[1]
+  assert 'function_responses' in call_args
+  assert call_args['function_responses'] == [function_response]
 
 
 @pytest.mark.asyncio