fix: separate thinking from text parts in streaming mode

Copybara import of the project:

--
79962881ca1c17eb6d7bd9dcf31a44df93c9badd by Almas Akchabayev <almas.akchabayev@gmail.com>:

fix: separate thinking from text parts in streaming mode
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/777 from almeynman:separate-thinking-and-text-parts-in-stream-mode b63dcc7fd0fc3973888dcbb9d4cc7e7e0a66e7f7
PiperOrigin-RevId: 764561932

tests/unittests/models/test_google_llm.py

@@ -206,6 +206,80 @@ async def test_generate_content_async_stream(gemini_llm, llm_request):
     mock_client.aio.models.generate_content_stream.assert_called_once()
 
 
+@pytest.mark.asyncio
+async def test_generate_content_async_stream_preserves_thinking_and_text_parts(
+    gemini_llm, llm_request
+):
+    with mock.patch.object(gemini_llm, "api_client") as mock_client:
+        class MockAsyncIterator:
+            def __init__(self, seq):
+                self._iter = iter(seq)
+
+            def __aiter__(self):
+                return self
+
+            async def __anext__(self):
+                try:
+                    return next(self._iter)
+                except StopIteration:
+                    raise StopAsyncIteration
+
+        response1 = types.GenerateContentResponse(
+            candidates=[
+                types.Candidate(
+                    content=Content(
+                        role="model",
+                        parts=[Part(text="Think1", thought=True)],
+                    ),
+                    finish_reason=None,
+                )
+            ]
+        )
+        response2 = types.GenerateContentResponse(
+            candidates=[
+                types.Candidate(
+                    content=Content(
+                        role="model",
+                        parts=[Part(text="Think2", thought=True)],
+                    ),
+                    finish_reason=None,
+                )
+            ]
+        )
+        response3 = types.GenerateContentResponse(
+            candidates=[
+                types.Candidate(
+                    content=Content(
+                        role="model",
+                        parts=[Part.from_text(text="Answer.")],
+                    ),
+                    finish_reason=types.FinishReason.STOP,
+                )
+            ]
+        )
+
+        async def mock_coro():
+            return MockAsyncIterator([response1, response2, response3])
+
+        mock_client.aio.models.generate_content_stream.return_value = mock_coro()
+
+        responses = [
+            resp
+            async for resp in gemini_llm.generate_content_async(
+                llm_request, stream=True
+            )
+        ]
+
+        assert len(responses) == 4
+        assert responses[0].partial is True
+        assert responses[1].partial is True
+        assert responses[2].partial is True
+        assert responses[3].content.parts[0].text == "Think1Think2"
+        assert responses[3].content.parts[0].thought is True
+        assert responses[3].content.parts[1].text == "Answer."
+        mock_client.aio.models.generate_content_stream.assert_called_once()
+
+
 @pytest.mark.asyncio
 async def test_connect(gemini_llm, llm_request):
   # Create a mock connection