Update test for provider-layer max_tokens normalization

After the oracle#148 refactor, OpenAIProvider.normalize_params remaps max_tokens
to max_completion_tokens at request-prep time. The agent layer now stays
model-agnostic and passes max_tokens through unchanged, so the test now
asserts that behavior instead of the old agents/common.py remap.

Author: fede-kamel

libs/oci/tests/unit_tests/agents/test_deep_research.py

@@ -531,8 +531,14 @@ def test_custom_model_id(self) -> None:
                     call_kwargs = mock_llm_class.call_args.kwargs
                     assert call_kwargs["model_id"] == "google.gemini-2.5-flash"
 
-    def test_openai_models_use_max_completion_tokens(self) -> None:
-        """Test OpenAI-compatible models map max_tokens to max_completion_tokens."""
+    def test_openai_models_pass_max_tokens_through(self) -> None:
+        """OpenAI-compatible models receive ``max_tokens`` from the agent layer.
+
+        The ``max_tokens -> max_completion_tokens`` remap happens at
+        request-prep time in ``OpenAIProvider.normalize_params`` (see
+        ``chat_models/providers/generic.py``), keeping the agent layer
+        model-agnostic.
+        """
         from langchain_oci.agents.deep_research.agent import create_deep_research_agent
 
         with patch.dict("os.environ", {"OCI_COMPARTMENT_ID": "test"}):
@@ -547,8 +553,8 @@ def test_openai_models_use_max_completion_tokens(self) -> None:
                     )
 
                     call_kwargs = mock_llm_class.call_args.kwargs
-                    assert call_kwargs["model_kwargs"]["max_completion_tokens"] == 1024
-                    assert "max_tokens" not in call_kwargs["model_kwargs"]
+                    assert call_kwargs["model_kwargs"]["max_tokens"] == 1024
+                    assert "max_completion_tokens" not in call_kwargs["model_kwargs"]
 
     def test_tools_passed_to_create_deep_agent(self) -> None:
         """Test that tools are passed to create_deep_agent."""