feat(factorminer): 新增 LocalFactorEvaluator 集成到评估管线

- 新增 LocalFactorEvaluator 类封装 FactorEngine，提供 (M,T) 矩阵输出
- evaluate_factors_with_evaluator() 支持新评估方式
- ValidationPipeline 优先使用 evaluator 计算信号
- 新增测试文件验证功能

tests/test_factorminer_e2e.py

@@ -0,0 +1,117 @@
+"""端到端集成测试：验证 110 个 Paper Factors 能被本地引擎计算。
+
+测试目标：
+1. 加载 paper factors 库
+2. 使用 LocalFactorEvaluator 计算每个因子
+3. 验证输出形状为 (M, T) 且不含全 NaN
+4. 统计成功率和跳过数
+
+运行命令：uv run pytest tests/test_factorminer_e2e.py -v
+"""
+
+import pytest
+
+
+def test_paper_factors_e2e():
+    """端到端测试：验证 110 个 Paper Factors 能被本地引擎计算。
+
+    注意：此测试需要数据库中有 pro_bar 数据。
+    如果数据不足，会在计算时失败。
+    """
+    from src.factorminer.core.library_io import import_from_paper
+    from src.factorminer.evaluation.local_engine import LocalFactorEvaluator
+
+    # 1. 加载 paper factors 库
+    library = import_from_paper()
+    factors = library.list_factors()
+    total = len(factors)
+
+    print(f"\n[e2e] 总计加载 {total} 个因子")
+
+    # 2. 过滤 unsupported 因子
+    supported_factors = [f for f in factors if not f.metadata.get("unsupported", False)]
+    unsupported_count = total - len(supported_factors)
+
+    print(f"[e2e] 可计算因子 {len(supported_factors)} 个")
+    print(f"[e2e] 跳过（未实现算子）{unsupported_count} 个")
+
+    if not supported_factors:
+        print("[e2e] 没有可计算的因子，测试跳过")
+        pytest.skip("No supported factors to test")
+
+    # 3. 实例化 LocalFactorEvaluator
+    evaluator = LocalFactorEvaluator(
+        start_date="20200101",
+        end_date="20201231",
+    )
+
+    # 4. 批量计算每个因子
+    success_count = 0
+    fail_count = 0
+    skip_count = 0
+
+    specs = [(f.name, f.formula) for f in supported_factors]
+
+    print(f"[e2e] 开始批量计算 {len(specs)} 个因子...")
+
+    try:
+        results = evaluator.evaluate(specs)
+    except Exception as e:
+        print(f"[e2e] 批量计算失败: {e}")
+        pytest.fail(f"Batch evaluation failed: {e}")
+
+    # 5. 验证结果
+    for name, result in results.items():
+        if result is None:
+            fail_count += 1
+            continue
+
+        # 检查形状
+        if result.ndim != 2:
+            print(f"[e2e] 因子 {name} 形状错误: {result.ndim}D（期望 2D）")
+            fail_count += 1
+            continue
+
+        # 检查是否全 NaN
+        if np.isnan(result).all():
+            print(f"[e2e] 因子 {name} 输出全为 NaN")
+            fail_count += 1
+            continue
+
+        success_count += 1
+
+    print(
+        f"[e2e] 成功 {success_count}/{total}，"
+        f"跳过 {skip_count} 个，失败 {fail_count} 个"
+    )
+
+    # 断言：至少 50% 的因子能成功计算（宽松标准）
+    success_rate = success_count / total
+    assert success_rate >= 0.5, f"成功率 {success_rate:.1%} 低于 50% 阈值"
+
+
+if __name__ == "__main__":
+    import numpy as np
+    from src.factorminer.core.library_io import import_from_paper
+    from src.factorminer.evaluation.local_engine import LocalFactorEvaluator
+
+    # 手动运行测试
+    library = import_from_paper()
+    factors = library.list_factors()
+    total = len(factors)
+    print(f"\n[e2e] 总计加载 {total} 个因子")
+
+    supported_factors = [f for f in factors if not f.metadata.get("unsupported", False)]
+    unsupported_count = total - len(supported_factors)
+    print(f"[e2e] 可计算因子 {len(supported_factors)} 个")
+    print(f"[e2e] 跳过（未实现算子）{unsupported_count} 个")
+
+    if supported_factors:
+        evaluator = LocalFactorEvaluator(
+            start_date="20200101",
+            end_date="20201231",
+        )
+        specs = [(f.name, f.formula) for f in supported_factors]
+        print(f"[e2e] 开始批量计算 {len(specs)} 个因子...")
+        results = evaluator.evaluate(specs)
+        print(f"[e2e] 批量计算完成，共 {len(results)} 个结果")