feat(rag): 子章节级图片过滤 + VLM 后台增强 + 意图分析优化

- 图片选择新增 section_path 字段，支持子章节级过滤
- _filter_images_by_answer 增加 primary_sections 参数和叶子节点匹配
- 检索发散检测：primary_leaf_names > 3 时全局阈值+1
- _FIGURE_ANSWER_KEYWORDS 移除单字"图""表"，正则图号兜底防误触发
- lazy_enhance 后台增强流程优化
- 意图分析与 LLM 工具层改进

评测：图片选择 F1 从 52.4% 提升至 66.3%（+13.9pp），Precision +16.5pp

knowledge/document.py

@@ -72,6 +72,18 @@ class DocumentMixin:
         except Exception as e:
             logger.warning(f"清理版本记录失败: {e}")
 
+        # 清理不再被引用的图片和 VLM 缓存文件
+        try:
+            from knowledge.image_cleanup import cleanup_image_orphans
+            cleanup_result = cleanup_image_orphans(self, collections=[kb_name])
+            if cleanup_result['deleted_images'] or cleanup_result['deleted_caches']:
+                logger.info(
+                    f"清理孤儿文件: {cleanup_result['deleted_images']} 图片 + "
+                    f"{cleanup_result['deleted_caches']} VLM缓存"
+                )
+        except Exception as e:
+            logger.warning(f"清理孤儿文件失败: {e}")
+
         logger.info(f"从 {kb_name} 删除文档: {filename}, 片段数: {deleted}")
         return deleted