llm_helpers: unwrap CreateResult.content defensively (fix Step 9b on gateway)

llm_call_expect_json did `result.content.content`, assuming a nested object.
On the OpenAI/phyagi-gateway path CreateResult.content is already a plain
str, so this raised "'str' object has no attribute 'content'" on every
attempt — Step 9b (trajectory-informed task verification) failed all retries
and silently fell back to a default. The bug was masked because the test only
asserts the result keys exist, not that the step ran.

Mirror task_classification's helper: unwrap .content only when present. Step
9b now runs clean against the gateway (no retries, real verdict).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

webeval/src/webeval/rubric_agent/llm_helpers.py

@@ -81,7 +81,13 @@ async def llm_call_expect_json(
             if json_output:
                 create_kwargs["json_output"] = True
             result = await client.create(**create_kwargs)
-            response_text = (result.content.content or "").strip()
+            # CreateResult.content is a plain str on the OpenAI/gateway path
+            # but a nested object on some Azure wrappers — unwrap defensively
+            # (mirrors task_classification's helper) so neither shape breaks.
+            content = result.content
+            if hasattr(content, "content"):
+                content = content.content
+            response_text = (content or "").strip()
         except Exception as e:
             # Don't nudge on transport errors — the LLM never saw the
             # prompt, so an "Error: ConnectionError" turn is just noise.