fix(image): handle ChatGPT new async image-gen response format

ChatGPT moved the `async_task_type=image_gen` marker off the actual
image node (it now only sits on the "正在处理" placeholder card), and
finished images arrive as a separate tool/multimodal_text message whose
metadata has no async_task_type and whose asset_pointer is a
`sediment://` ref. The old strict gate
`metadata.async_task_type != "image_gen"` skipped the real image node,
so _poll_image_results always timed out -> "image task returned no
image data".

Drop the async_task_type gate; collect any tool multimodal_text message
that carries file-service:// or sediment:// pointers, and skip messages
with no image pointers. Handles both old and new formats, and now
returns both images of the A/B (paragen) pair.

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

services/openai_backend_api.py

@@ -619,10 +619,12 @@ class OpenAIBackendAPI:
             content = message.get("content") or {}
             if author.get("role") != "tool":
                 continue
-            if metadata.get("async_task_type") != "image_gen":
-                continue
             if content.get("content_type") != "multimodal_text":
                 continue
+            # 新版异步生图: 图片节点不再带 metadata.async_task_type=image_gen,
+            # 该标记只留在"正在处理"占位卡上。这里改为只要是 tool 的 multimodal_text
+            # 且 parts 里含图片指针(file-service:// 或 sediment://)就采集。
+            _ = metadata  # async_task_type 闸门已废弃,见上
             file_ids, sediment_ids = [], []
             for part in content.get("parts") or []:
                 text = (part.get("asset_pointer") or "") if isinstance(part, dict) else (
@@ -633,6 +635,8 @@ class OpenAIBackendAPI:
                 for hit in sed_pat.findall(text):
                     if hit not in sediment_ids:
                         sediment_ids.append(hit)
+            if not file_ids and not sediment_ids:
+                continue
             records.append(
                 {"message_id": message_id, "create_time": message.get("create_time") or 0, "file_ids": file_ids,
                  "sediment_ids": sediment_ids})