fix(openai): re-export NamedWrapper from braintrust.oai for back-compat (#319)

autoevals imports NamedWrapper from braintrust.oai to detect whether an
OpenAI client has already been wrapped by the Braintrust SDK. After the
0.15.0 refactor moved the class to
braintrust.integrations.openai.tracing,
the import silently fell back to an internal stub, so isinstance checks
no longer matched braintrust-wrapped clients. autoevals then re-wrapped
with its own non-tracing shim, dropping scorer spans
(openai.responses.create
and Embedding calls with purpose="scorer") across a number of expect
tests.

Re-export NamedWrapper from braintrust.oai so autoevals' existing
detection
path keeps working.

Author: AbhiPrasad

py/src/braintrust/oai.py

@@ -1,12 +1,20 @@
 from braintrust.integrations.openai import OpenAIIntegration, wrap_openai
 
+# NamedWrapper is re-exported here for back-compat. autoevals imports it from
+# `braintrust.oai` to detect OpenAI clients that the Braintrust SDK has already
+# wrapped (see autoevals/oai.py); if the import fails, autoevals falls back to
+# an internal stub and its isinstance check misses real wrapped clients,
+# causing scorer LLM calls to be re-wrapped without tracing.
+from braintrust.integrations.openai.tracing import NamedWrapper
+
 
 def patch_openai() -> bool:
     """Patch OpenAI globally for Braintrust tracing."""
     return OpenAIIntegration.setup()
 
 
 __all__ = [
+    "NamedWrapper",
     "patch_openai",
     "wrap_openai",
 ]