fix(vision): unwrap fast_json list-wrapped object

apps/worker/worker_app/tasks/vision/vlm_fallback.py

@@ -141,11 +141,11 @@ def call_pro_vlm(
             usage.get("completion_tokens", 0),
         )
         text = _strip_code_fence(raw)
-        l, r_pos = text.find("{"), text.rfind("}")
-        if l < 0 or r_pos <= l:
+        lb, r_pos = text.find("{"), text.rfind("}")
+        if lb < 0 or r_pos <= lb:
             logger.warning("[vision.v2] pro 无JSON elapsed=%.1fs head=%s", elapsed, raw[:200])
             return None
-        obj = json.loads(text[l : r_pos + 1])
+        obj = json.loads(text[lb : r_pos + 1])
         if not isinstance(obj, dict):
             return None
         scene = obj.get("scene") or "通用"

apps/worker/worker_app/tasks/vision/vlm_fast_json.py

@@ -153,14 +153,21 @@ def call_fast_json(
             reasoning_tokens,
         )
         text = _strip_code_fence(raw)
-        l, r = text.find("{"), text.rfind("}")
-        if l >= 0 and r > l:
-            text = text[l : r + 1]
+        lb, rb = text.find("{"), text.rfind("}")
+        if lb >= 0 and rb > lb:
+            text = text[lb : rb + 1]
         try:
             obj = json.loads(text)
         except json.JSONDecodeError:
             logger.warning("[vision.v2] fast_json JSON 解析失败 elapsed=%.1fs head=%s", elapsed, raw[:200])
             return None
+        if isinstance(obj, list):
+            # 模型偶发将对象包在数组里，取第一个 dict 元素
+            dicts = [x for x in obj if isinstance(x, dict)]
+            if not dicts:
+                logger.warning("[vision.v2] fast_json list 内无 dict: %s", type(obj))
+                return None
+            obj = dicts[0]
         if not isinstance(obj, dict):
             logger.warning("[vision.v2] fast_json 非 dict: %s", type(obj))
             return None