feat(server-release): 从 main 同步切片层级结构修复与解析增强

- parsers/mineru_parser.py: MinerU V2 解析增强 + 切片层级结构修复
- parsers/heading_rules.py: 标题识别规则引擎增强
- knowledge/manager.py: 跨页表格合并与切片后处理优化

切片数从旧版 819 降至 437（与本地 VLM 解析的 448 接近）

knowledge/manager.py

@@ -139,7 +139,7 @@ class KnowledgeBaseManager(
         logger.info(f"知识库管理器初始化完成，路径: {self.base_path}，发现 {len(existing_kbs)} 个向量库: {existing_kbs}")
 
     def _load_metadata(self) -> dict:
-        """加载元数据（带文件锁）"""
+        """加载元数据（带文件锁，确保多 worker 进程间一致）"""
         metadata_path = os.path.join(self.base_path, KB_METADATA_FILE)
         if os.path.exists(metadata_path):
             try:
@@ -156,7 +156,7 @@ class KnowledgeBaseManager(
         return {"collections": {}}
 
     def _save_metadata(self):
-        """保存元数据（带文件锁，防止并发写入覆盖）"""
+        """保存元数据（带文件锁，防止并发写入数据覆盖）"""
         metadata_path = os.path.join(self.base_path, KB_METADATA_FILE)
         try:
             with open(metadata_path, 'w', encoding='utf-8') as f:
@@ -335,6 +335,7 @@ class KnowledgeBaseManager(
                 "section": section_path,
                 "status": "active",
                 "version": "v1",
+                "text_level": getattr(chunk, 'text_level', 0),  # 标题级别(0=正文,1=h1,2=h2,3=h3)，供检索管线层次感知
             }
 
             if extra_metadata:
@@ -347,6 +348,25 @@ class KnowledgeBaseManager(
             if hasattr(chunk, 'image_path') and chunk.image_path:
                 metadata['image_path'] = chunk.image_path
 
+            # bbox 坐标（PDF 有，DOCX 无，需 None 保护）
+            # _build_citation() 从 metadata 读取 bbox 做引用定位
+            chunk_bbox = getattr(chunk, 'bbox', None)
+            if chunk_bbox:
+                metadata['bbox'] = json.dumps(chunk_bbox)
+
+            # MinerU 结构化元数据（表格类型、嵌套层级、图片子类型）
+            chunk_table_type = getattr(chunk, 'table_type', '')
+            if chunk_table_type:
+                metadata['table_type'] = chunk_table_type
+
+            chunk_nest_level = getattr(chunk, 'table_nest_level', '')
+            if chunk_nest_level:
+                metadata['table_nest_level'] = str(chunk_nest_level)
+
+            chunk_sub_type = getattr(chunk, 'sub_type', '')
+            if chunk_sub_type:
+                metadata['sub_type'] = chunk_sub_type
+
             # 生成向量
             try:
                 embedding = embedding_model.encode(semantic_content).tolist()
@@ -602,7 +622,7 @@ class KnowledgeBaseManager(
         try:
             from config import get_llm_client, DASHSCOPE_MODEL
             client = get_llm_client()
-            summary = call_llm(client, prompt, DASHSCOPE_MODEL, max_tokens=2048)
+            summary = call_llm(client, prompt, DASHSCOPE_MODEL, max_tokens=512)
             return summary.strip() if summary else ""
         except Exception as e:
             logger.warning(f"生成表格摘要失败: {e}")
@@ -661,31 +681,13 @@ class KnowledgeBaseManager(
                         ]
                     }
                 ],
-                max_tokens=2048  # mimo-v2.5 推理模型思考链消耗 ~1000 token，需留足输出空间
+                max_tokens=512
             )
 
             description = response.choices[0].message.content
 
-            # 推理模型兼容：content 为空时从 reasoning_content 提取
-            if not description or not description.strip():
-                reasoning = getattr(response.choices[0].message, 'reasoning_content', None)
-                if reasoning and reasoning.strip():
-                    import re
-                    # 尝试从思考链中提取有用文本（去掉 <think> 标签后的内容）
-                    cleaned = re.sub(r'', '', reasoning, flags=re.DOTALL).strip()
-                    if cleaned:
-                        logger.info(f"VLM content为空，从reasoning_content提取描述: {image_path}")
-                        description = cleaned
-                    else:
-                        description = reasoning.strip()
-
-            if not description:
-                logger.warning(f"VLM 返回空描述: {image_path}")
-                return ""
-
             # 缓存结果
             import hashlib
-            import re as _re
             img_hash = hashlib.md5(img_path.read_bytes()).hexdigest()
             cache_dir = Path('.data/cache/vlm')
             cache_dir.mkdir(parents=True, exist_ok=True)