test: move the profile-hook keyword compat test into its own module

test_chat_completions.py grew past the 1,000-line test-file guideline with #134320's
TestProfileHookKeywordCompat (1,006 lines). The test pins the transport's hook-call
contract, not response handling, so it moves verbatim to test_profile_hook_keyword_compat.py
(985 lines left). Reported by the review bot on #134320.

tests/agent/transports/test_chat_completions.py

@@ -546,27 +546,6 @@ class TestChatCompletionsLmStudioReasoning:
         assert kw["reasoning_effort"] == "high"
 
 
-
-
-
-class TestProfileHookKeywordCompat:
-    """``lmstudio_reasoning_options`` is LM Studio-only: a plugin hook that names the older keywords
-    without ``**context`` must not start raising TypeError on every request."""
-
-    def test_hook_without_var_kwargs_still_called(self, transport):
-        from providers.base import ProviderProfile
-
-        class OldSignature(ProviderProfile):
-            def build_api_kwargs_extras(self, *, reasoning_config=None, supports_reasoning=False,
-                                        qwen_session_metadata=None, model=None, base_url=None,
-                                        ollama_num_ctx=None, session_id=None, cache_scope_id=None):
-                return {}, {"reasoning_effort": "low"}
-
-        kw = transport.build_kwargs(model="m", messages=[{"role": "user", "content": "Hi"}],
-                                    provider_profile=OldSignature(name="old-signature"))
-        assert kw["reasoning_effort"] == "low"
-
-
 class TestChatCompletionsValidate:
 
     def test_none(self, transport):

tests/agent/transports/test_profile_hook_keyword_compat.py

@@ -0,0 +1,33 @@
+"""A plugin profile's ``build_api_kwargs_extras`` written against the older keyword set keeps working.
+
+Split from ``test_chat_completions.py`` (size cap): this pins the transport's hook-call contract,
+not response handling.
+"""
+
+import pytest
+
+from agent.transports import get_transport
+
+
+@pytest.fixture
+def transport():
+    import agent.transports.chat_completions  # noqa: F401
+    return get_transport("chat_completions")
+
+
+class TestProfileHookKeywordCompat:
+    """``lmstudio_reasoning_options`` is LM Studio-only: a plugin hook that names the older keywords
+    without ``**context`` must not start raising TypeError on every request."""
+
+    def test_hook_without_var_kwargs_still_called(self, transport):
+        from providers.base import ProviderProfile
+
+        class OldSignature(ProviderProfile):
+            def build_api_kwargs_extras(self, *, reasoning_config=None, supports_reasoning=False,
+                                        qwen_session_metadata=None, model=None, base_url=None,
+                                        ollama_num_ctx=None, session_id=None, cache_scope_id=None):
+                return {}, {"reasoning_effort": "low"}
+
+        kw = transport.build_kwargs(model="m", messages=[{"role": "user", "content": "Hi"}],
+                                    provider_profile=OldSignature(name="old-signature"))
+        assert kw["reasoning_effort"] == "low"