fix: sanitize OpenAI tracing export payloads (#2896)

### Summary

Updates the OpenAI tracing exporter to align usage metadata with the
trace ingest payload shape.

For the default OpenAI tracing endpoint, usage metadata is now only
included on generation spans, where it is normalized to the supported
token-count shape. Other tracing endpoints continue to receive the raw
SDK payload.

Adds regression coverage for OpenAI endpoint sanitization and custom
endpoint behavior.

### Test plan

- `make format`
- `make lint`
- `make sync && make typecheck`
- `uv run pytest
tests/test_openai_responses.py::test_get_response_span_exports_usage
tests/test_trace_processor.py` (46 passed)
- Manual tracing repro: `Runner.run(...)` completed, `flush_traces()`
completed, and no tracing client errors were emitted.
- `make tests` reached 3858 passed, 10 failed locally in sandbox tests
because `sandbox-exec` returned `sandbox_apply: Operation not
permitted`.

### Issue number

N/A

### Checks

- [x] I've added new tests (if relevant)
- [ ] I've added/updated the relevant documentation
- [x] I've run `make lint` and `make format`
- [ ] I've made sure tests pass

Author: alfozan

src/agents/tracing/processors.py

@@ -39,6 +39,7 @@ class BackendSpanExporter(TracingExporter):
             "output_tokens",
         }
     )
+    _OPENAI_TRACING_USAGE_SPAN_TYPES = frozenset({"generation"})
     _UNSERIALIZABLE = object()
 
     def __init__(
@@ -203,7 +204,12 @@ def _sanitize_for_openai_tracing_api(self, payload_item: dict[str, Any]) -> dict
                 did_mutate = True
             sanitized_span_data[field_name] = sanitized_field
 
-        if span_data.get("type") != "generation":
+        if span_data.get("type") not in self._OPENAI_TRACING_USAGE_SPAN_TYPES:
+            if "usage" in span_data:
+                if not did_mutate:
+                    sanitized_span_data = dict(span_data)
+                    did_mutate = True
+                sanitized_span_data.pop("usage", None)
             if not did_mutate:
                 return payload_item
             sanitized_payload_item = dict(payload_item)

tests/test_trace_processor.py

@@ -572,7 +572,7 @@ def export(self):
 
 
 @patch("httpx.Client")
-def test_backend_span_exporter_does_not_modify_non_generation_usage(mock_client):
+def test_backend_span_exporter_drops_non_generation_usage_for_openai_endpoint(mock_client):
     class DummyItem:
         tracing_api_key = None
 
@@ -592,6 +592,35 @@ def export(self):
     exporter = BackendSpanExporter(api_key="test_key")
     exporter.export([cast(Any, DummyItem())])
 
+    sent_payload = mock_client.return_value.post.call_args.kwargs["json"]["data"][0]
+    assert "usage" not in sent_payload["span_data"]
+    exporter.close()
+
+
+@patch("httpx.Client")
+def test_backend_span_exporter_keeps_non_generation_usage_for_custom_endpoint(mock_client):
+    class DummyItem:
+        tracing_api_key = None
+
+        def export(self):
+            return {
+                "object": "trace.span",
+                "span_data": {
+                    "type": "function",
+                    "usage": {"requests": 1},
+                },
+            }
+
+    mock_response = MagicMock()
+    mock_response.status_code = 200
+    mock_client.return_value.post.return_value = mock_response
+
+    exporter = BackendSpanExporter(
+        api_key="test_key",
+        endpoint="https://example.com/v1/traces/ingest",
+    )
+    exporter.export([cast(Any, DummyItem())])
+
     sent_payload = mock_client.return_value.post.call_args.kwargs["json"]["data"][0]
     assert sent_payload["span_data"]["usage"] == {"requests": 1}
     exporter.close()