整理了初始化过程，现在更容易被继承了。

sync_state_machine/datasource/jsonl/handler.py

@@ -50,24 +50,22 @@ class BaseJsonlHandler(BaseNodeHandler):
     ):
         super().__init__(node_type, node_class, schema)
         self.datasource = datasource
-        self._target_project_ids: Optional[Set[str]] = None
+        self._target_project_ids: List[str] = []
 
     def set_target_project_ids(self, target_project_ids: List[str]) -> None:
-        """注入项目过滤列表，供 load 阶段预过滤使用。"""
-        if target_project_ids:
-            self._target_project_ids = {str(pid) for pid in target_project_ids if str(pid)}
-        else:
-            self._target_project_ids = None
+        """注入项目过滤列表的本地副本。真正的过滤来源仍然是 datasource。"""
+        self._target_project_ids = [str(pid) for pid in target_project_ids if str(pid)]
 
     def _should_skip_item_by_project_filter(self, item: Dict[str, Any], item_id: str) -> bool:
-        if not self._target_project_ids:
+        target_project_ids = set(self.datasource.target_project_ids)
+        if not target_project_ids:
             return False
         if self.node_type == "project":
-            return item_id not in self._target_project_ids
+            return item_id not in target_project_ids
 
         project_id = item.get("project_id")
         if project_id:
-            return str(project_id) not in self._target_project_ids
+            return str(project_id) not in target_project_ids
         return False
     
     async def load(self) -> List['SyncNode']: