fix(tests): use litellm.anthropic_messages entrypoint + drop unstable openapi field

Two CI failures, both pre-existing in different ways:

1. reasoning_effort_grid: all 33 bedrock_invoke_messages cells failed with
   AttributeError("module 'litellm' has no attribute 'messages'"). litellm
   exposes the async Anthropic Messages entrypoint as litellm.anthropic_messages
   (via "from .llms.anthropic.experimental_pass_through.messages.handler
   import *" in litellm/__init__.py), not litellm.messages.acreate. Swap
   the call.

2. tests/test_litellm/interactions/test_openapi_compliance.py::TestResponseCompliance::test_interaction_response_fields
   asserts the live Google spec contains "steps". Google's spec has churned
   through "outputs" -> "steps" -> neither, and presently carries neither.
   The test broke on main as soon as upstream dropped "steps"; pulling the
   key off the assert list realigns the test with the live schema. Re-add
   the per-turn output field once upstream stabilizes on a name.

The openapi-compliance fix doesn't belong to this PR conceptually but is
included here per request to unblock CI before the morning.

tests/llm_translation/reasoning_effort_grid/test_reasoning_effort_grid.py

@@ -172,7 +172,7 @@ async def _call_messages(
 ) -> Tuple[int, Optional[Exception]]:
     kwargs = _build_messages_kwargs(model, effort)
     try:
-        await litellm.messages.acreate(**kwargs)
+        await litellm.anthropic_messages(**kwargs)
         return 200, None
     except BadRequestError as exc:
         return 400, exc

tests/test_litellm/interactions/test_openapi_compliance.py

@@ -157,15 +157,17 @@ class TestResponseCompliance:
         # Check CreateModelInteractionParams which includes output fields
         schema = spec_dict["components"]["schemas"]["CreateModelInteractionParams"]
 
-        # Output fields (readOnly). Google renamed `outputs` → `steps` in the
-        # upstream spec; keep this list aligned with the live schema.
+        # Output fields (readOnly). Google's live spec has churned through
+        # both `outputs` and `steps` for the per-turn output array and at the
+        # moment carries neither -- only the stable response-level fields
+        # below are guaranteed. Re-add the per-turn key once upstream
+        # stabilizes on a name.
         output_fields = [
             "id",
             "status",
             "created",
             "updated",
             "role",
-            "steps",
             "usage",
         ]