feat: 生图模型可选(火山/gpt-image) + 工作台图「加入资产库」收纳 + 任务中心排序修复

- AI 工作室生图支持显式选模型:resolve_image_model 解析 volcano(Seedream 图生图,
  无 image_edit 时走 image_generation 带参考图)/ gpt-image(image_edit 多图编辑),
  未选回落系统默认;enqueue_standalone_images 透传 image_model
- 资产库收纳:Asset 加 in_library 字段(迁移 0007,既有资产 db_default=True 不动);
  工作台生成图默认 in_library=False,只在工作台展示,用户「加入资产库」后才进库列表;
  assets 视图/序列化器/library 页/types 配套
- 任务中心修复:AITaskViewSet 默认 order_by(-created_at),前端 aiTasks 取 page_size=200。
  原因:列表无排序时 MySQL 按 UUID 主键乱序返回,把一批旧失败记录顶到首页,
  前端又只取首页 20 条算 全部/已完成/失败 → 误显示「全部失败」(实为 421 成功/少量失败)

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

core/backend/apps/ai/views.py

@@ -34,9 +34,10 @@ class GenerateImageView(APIView):
         model_id = str(request.data.get("model_id") or "").strip() or None
         model_entity_id = str(request.data.get("model_entity_id") or "").strip() or None
         ratio = str(request.data.get("ratio") or "").strip() or None
+        image_model = str(request.data.get("image_model") or "").strip() or None
         team = get_current_team(request.user)
         try:
-            tasks = enqueue_standalone_images(team=team, user=request.user, prompt=prompt, mode=mode, count=count, product_id=product_id, reference_product=reference_product, model_id=model_id, model_entity_id=model_entity_id, ratio=ratio)
+            tasks = enqueue_standalone_images(team=team, user=request.user, prompt=prompt, mode=mode, count=count, product_id=product_id, reference_product=reference_product, model_id=model_id, model_entity_id=model_entity_id, ratio=ratio, image_model=image_model)
         except ValueError as exc:  # 无可用模型 / 余额不足等,立即反馈
             return Response({"detail": str(exc)}, status=status.HTTP_400_BAD_REQUEST)
         return Response(
@@ -69,7 +70,10 @@ class GenerateImageView(APIView):
 class AITaskViewSet(TeamScopedViewSetMixin, ReadOnlyModelViewSet):
     # 序列化器不含 request_payload/response_payload(单条可达 3MB+ base64 图),defer 掉:
     # 否则只为序列化 14 个小字段也会把几十 MB blob 从库里拉回(远程库实测 40 条要 30s+)。
-    queryset = AITask.objects.select_related("team", "project", "model_config", "model_config__provider").defer("request_payload", "response_payload").all()
+    # 默认按创建时间倒序:任务中心 = 历史流水,最新的(多为成功)排最前。
+    # 缺省排序时 MySQL 按主键(UUID)乱序返回,会把一批旧失败记录顶到首页,
+    # 前端只取首页 → 误判「全部失败」。order_by 保证稳定且新任务优先。
+    queryset = AITask.objects.select_related("team", "project", "model_config", "model_config__provider").defer("request_payload", "response_payload").order_by("-created_at")
     serializer_class = AITaskSerializer
     search_fields = ["idempotency_key", "provider_task_id", "project__name"]
     ordering_fields = ["created_at", "updated_at", "completed_at"]