test(ci): serve /moderations from the canned OpenAI mock (#37739)

* test(ci): serve /moderations from the canned OpenAI mock

The otel proxy E2E job points its `openai/*` wildcard deployment at the
canned mock, and #37492 made `get_model_list` agree with
`get_available_deployment` on bare model names. /moderations now resolves
`omni-moderation-latest` to that wildcard deployment the way
/chat/completions already did, so the request lands on the mock, which
never implemented the route and answers a bare 404.

Add /moderations and /v1/moderations to the mock, returning an
OpenAI-shaped response with one result per input item.

* style(ci): annotate the new moderations locals as Final

tests/_fake_openai_endpoint_server.py

@@ -8,11 +8,11 @@ those jobs failed with ``404 Application not found`` even though nothing in the
 PR was broken.
 
 This process is the local stand-in. A model points its ``api_base`` here and
-gets back a well-formed chat/text/embedding response with realistic ``usage`` so
+gets back a well-formed chat/text/embedding/moderation response with realistic
-cost tracking and spend accounting still exercise their real code paths. The one
+``usage`` so cost tracking and spend accounting still exercise their real code
-behavioral special case mirrors the old hosted mock: a request whose ``model``
+paths. The one behavioral special case mirrors the old hosted mock: a request
-is ``429`` returns HTTP 429 so rate-limit and cooldown tests still have
+whose ``model`` is ``429`` returns HTTP 429 so rate-limit and cooldown tests
-something to trip on.
+still have something to trip on.
 """
 
 from __future__ import annotations
@@ -35,6 +35,21 @@ _SLOW_MODEL: Final = "slow-endpoint"
 _SLOW_RESPONSE_SECONDS: Final = 3.0
 _PROMPT_TOKENS: Final = 20
 _COMPLETION_TOKENS: Final = 20
+_MODERATION_CATEGORIES: Final = (
+    "harassment",
+    "harassment/threatening",
+    "hate",
+    "hate/threatening",
+    "illicit",
+    "illicit/violent",
+    "self-harm",
+    "self-harm/instructions",
+    "self-harm/intent",
+    "sexual",
+    "sexual/minors",
+    "violence",
+    "violence/graphic",
+)
 
 
 def _usage() -> dict[str, int]:
@@ -220,6 +235,28 @@ async def triton_embeddings(_request: Request) -> Response:
     )
 
 
+def _moderation_result() -> dict[str, object]:
+    return {
+        "flagged": False,
+        "categories": {category: False for category in _MODERATION_CATEGORIES},
+        "category_scores": {category: 0.0 for category in _MODERATION_CATEGORIES},
+        "category_applied_input_types": {category: ["text"] for category in _MODERATION_CATEGORIES},
+    }
+
+
+async def moderations(request: Request) -> Response:
+    body: Final = await _parse_body(request)
+    raw_input: Final = body.get("input", "")
+    count: Final = len(raw_input) if isinstance(raw_input, list) else 1
+    return JSONResponse(
+        {
+            "id": f"modr-{uuid.uuid4().hex[:24]}",
+            "model": _requested_model(body),
+            "results": [_moderation_result() for _ in range(max(count, 1))],
+        }
+    )
+
+
 async def list_models(_request: Request) -> Response:
     return JSONResponse(
         {
@@ -247,6 +284,8 @@ app = Starlette(
         Route("/embeddings", embeddings, methods=["POST"]),
         Route("/v1/embeddings", embeddings, methods=["POST"]),
         Route("/triton/embeddings", triton_embeddings, methods=["POST"]),
+        Route("/moderations", moderations, methods=["POST"]),
+        Route("/v1/moderations", moderations, methods=["POST"]),
         Route("/models", list_models, methods=["GET"]),
         Route("/v1/models", list_models, methods=["GET"]),
     ]

tests/local_testing/test_fake_openai_endpoint.py

@@ -13,9 +13,11 @@ from __future__ import annotations
 
 import re
 from pathlib import Path
+from typing import Final
 
 import httpx
 import pytest
+from openai.types import ModerationCreateResponse
 
 from tests.fake_openai_endpoint import (
     _LOCAL_DEFAULT,
@@ -56,6 +58,20 @@ def test_chat_completion_shape():
     assert body["usage"]["total_tokens"] == 40
 
 
+def test_moderations_route_parses_as_an_openai_response():
+    base: Final = ensure_fake_openai_endpoint()
+    response: Final = httpx.post(
+        f"{base}/v1/moderations",
+        json={"input": ["I want to harm someone", "hello"], "model": "omni-moderation-latest"},
+        timeout=10,
+    )
+    assert response.status_code == 200
+    parsed: Final = ModerationCreateResponse.model_validate(response.json())
+    assert parsed.model == "omni-moderation-latest"
+    assert len(parsed.results) == 2
+    assert parsed.results[0].categories.violence is False
+
+
 def test_triton_embeddings_route():
     base = ensure_fake_openai_endpoint()
     response = httpx.post(f"{base}/triton/embeddings", json={"inputs": []}, timeout=10)