Python: Fix OpenAIEmbeddingClient to use AsyncOpenAI for /openai/v1 endpoints (#5137)

* Fix OpenAIEmbeddingClient with /openai/v1 endpoint (#5068)

When base_url ends with /openai/v1/ and a credential is provided,
load_openai_service_settings was creating an AsyncAzureOpenAI client.
The Azure SDK rewrites deployment-based endpoints (including /embeddings)
by inserting /deployments/{model}/ into the URL, producing 404s on the
OpenAI-compatible /openai/v1 endpoint.

Use AsyncOpenAI instead of AsyncAzureOpenAI when the resolved base_url
targets /openai/v1, converting the Azure token provider to an async
api_key callable. The responses_mode path is unaffected because the
Responses API (/responses) is not in the SDK's rewrite list.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Python: Fix OpenAIEmbeddingClient to use AsyncOpenAI for /openai/v1 endpoints

Fixes #5068

* Address review feedback: improve test coverage and remove unrelated changes

- Revert unrelated formatting change in test_a2a_agent.py
- Fix test_init_with_openai_v1_base_url_and_api_key_uses_openai_client to
  exercise the Azure settings path (via AZURE_OPENAI_BASE_URL env var)
  instead of the plain OpenAI path, covering the elif api_key branch
- Add _ensure_async_token_provider unit tests for both sync and async
  token providers

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Address review feedback for #5068: Python: [Bug]: `OpenAIEmbeddingClient` does not work with `/openai/v1` endpoint

---------

Co-authored-by: MAF Dashboard Bot <maf-dashboard-bot@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Evan Mattson <35585003+moonbox3@users.noreply.github.com>

python/packages/openai/agent_framework_openai/_shared.py

@@ -282,9 +282,46 @@ def load_openai_service_settings(
             "Azure OpenAI client requires either an API key or an Azure AD token provider."
             " This can be provided either as a callable api_key or via the credential parameter."
         )
+
+    # The /openai/v1 endpoint exposes an OpenAI-compatible API surface.
+    # AsyncAzureOpenAI rewrites certain request paths (e.g. /embeddings,
+    # /chat/completions) by inserting /deployments/{model}/, which produces
+    # 404s on this endpoint.  Use AsyncOpenAI instead so request URLs are
+    # sent as-is.  responses_mode is excluded because the Responses API path
+    # (/responses) is not rewritten by the Azure SDK.
+    resolved_base_url = client_args.get("base_url", "")
+    if not responses_mode and resolved_base_url and resolved_base_url.rstrip("/").endswith("/openai/v1"):
+        openai_args: dict[str, Any] = {
+            "base_url": resolved_base_url,
+            "default_headers": client_args.get("default_headers"),
+        }
+        if "azure_ad_token_provider" in client_args:
+            openai_args["api_key"] = _ensure_async_token_provider(client_args["azure_ad_token_provider"])
+        elif "api_key" in client_args:
+            openai_args["api_key"] = client_args["api_key"]
+        return azure_settings, AsyncOpenAI(**openai_args), True  # type: ignore[return-value]
+
     return azure_settings, AsyncAzureOpenAI(**client_args), True  # type: ignore[return-value]
 
 
+def _ensure_async_token_provider(
+    provider: AzureTokenProvider,
+) -> Callable[[], Awaitable[str]]:
+    """Wrap a (possibly synchronous) token provider so it always returns an awaitable.
+
+    ``AsyncOpenAI`` requires callable ``api_key`` values to return ``Awaitable[str]``.
+    Azure token providers may return a plain ``str``, so this normalises them.
+    """
+
+    async def _wrapper() -> str:
+        result = provider()
+        if isinstance(result, str):
+            return result
+        return await result
+
+    return _wrapper
+
+
 def _resolve_azure_credential_to_token_provider(
     credential: AzureCredentialTypes | AzureTokenProvider,
 ) -> AzureTokenProvider: