refactor: 重构RAG核心组件，简化代码结构和测试文件

rag_indexer/index_builder.py

@@ -45,6 +45,11 @@ class IndexBuilderConfig:
     child_chunk_size: int = 200
     child_chunk_overlap: int = 20
     child_splitter_type: SplitterType = SplitterType.SEMANTIC   # 子块默认语义切分
+    # 子块语义切分参数
+    child_buffer_size: int = 1
+    child_breakpoint_threshold_type: str = "percentile"
+    child_breakpoint_threshold_amount: float = 90  # 降低阈值，让切分更激进
+    child_min_chunk_size: int = 50  # 降低最小块大小
 
     # 检索参数
     search_k: int = 5
@@ -86,7 +91,6 @@ class IndexBuilder:
         # 初始化向量存储（自动支持稠密+稀疏混合检索）
         self.vector_store = QdrantHybridStore(
             collection_name=config.collection_name,
-            embeddings=self.embeddings,
         )
         logger.info("✅ 混合检索向量存储初始化成功（稠密+BM25稀疏）")
 
@@ -125,6 +129,10 @@ class IndexBuilder:
             self.child_splitter = get_splitter(
                 SplitterType.SEMANTIC,
                 embeddings=self.embeddings,
+                buffer_size=cfg.child_buffer_size,
+                breakpoint_threshold_type=cfg.child_breakpoint_threshold_type,
+                breakpoint_threshold_amount=cfg.child_breakpoint_threshold_amount,
+                min_chunk_size=cfg.child_min_chunk_size,
                 **cfg.extra_splitter_kwargs
             )
         else: