feat: 量化引擎加固 — 新增测试 + 数据/因子/回测层优化

- 新增 finance/tests/ 6 个测试套件（agents/backtest/dao_upsert/factors/features/fundamental_lookahead）
- 数据层: data_manager / dao 优化，新增 upsert 逻辑
- 因子层: 基本面因子抽象定位 _mapping、ROE/PE/PB 重构
- 回测层: vectorbt/engine 大改动（251 行），report 增强
- ML 层: features/backtest_integration 特征工程与回测优化
- CLI: agent_cli 重构
- config/settings 扩充配置项

finance/tests/test_agents.py

@@ -0,0 +1,87 @@
+"""
+Agent 层回归测试：RiskAgent 配置集中可用性 + Orchestrator 每日流程步骤隔离。
+"""
+import sys
+import os
+sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
+
+import numpy as np
+import pandas as pd
+import pytest
+from unittest.mock import patch
+
+from agents.risk_agent import RiskAgent
+from agents.orchestrator import AgentOrchestrator, TOTAL_STEPS
+
+
+def _flat_dm():
+    """给出平淡行情 → 风险 low。"""
+    class _DM:
+        def get_daily(self, code):
+            idx = pd.date_range('2025-01-01', periods=300, freq='B').strftime('%Y%m%d')
+            close = np.linspace(100, 100 + 0.03 * 300, 300) + np.random.default_rng(0).normal(0, 0.2, 300)
+            return pd.DataFrame({'trade_date': idx, 'close': close})
+    return _DM()
+
+
+def _hot_dm():
+    class _DM:
+        def get_daily(self, code):
+            idx = pd.date_range('2025-01-01', periods=300, freq='B').strftime('%Y%m%d')
+            close = 100 * np.exp(np.cumsum(np.random.default_rng(1).normal(0, 0.03, 300)))
+            return pd.DataFrame({'trade_date': idx, 'close': close})
+    return _DM()
+
+
+def test_risk_config_low_level():
+    r = RiskAgent(dm=_flat_dm()).execute()
+    assert r['risk_level'] in ('low', 'medium', 'high')
+    assert r['target_exposure'] in (0.85, 0.60, 0.30)
+
+
+def test_risk_config_high_level():
+    r = RiskAgent(dm=_hot_dm()).execute()
+    assert r['risk_level'] in ('high', 'medium')
+
+
+class _FailAgent:
+    def execute(self, **kw):
+        raise RuntimeError("boom")
+
+
+class _OkAgent:
+    def execute(self, **kw):
+        return {"report_path": "/tmp/x.md", "risk_level": "low",
+                "target_exposure": 0.8, "top_picks": [], "date": "20260601"}
+
+
+class _FakeDM:
+    def get_stock_list(self):
+        return pd.DataFrame(index=['000001.SZ', '600519.SH'])
+    def sync_daily(self, code):
+        return 5
+
+
+class _FakeSent:
+    def get_scope_stocks(self):
+        return ["000001.SZ", "600519.SH"]
+    def compute(self, *a, **k):
+        return pd.DataFrame({"news_sent_5": [0.1, 0.2]})
+
+
+def test_run_daily_steps_isolated():
+    """risk/selection 抛错被隔离，report/sentiment/sync 仍成功。"""
+    with patch('database.dao.get_latest_trade_date', side_effect=lambda c: "20260530"):
+        orch = AgentOrchestrator(dm=_FakeDM(), sent=_FakeSent())
+        orch.agents = {"risk": _FailAgent(), "selection": _FailAgent(),
+                       "report": _OkAgent(), "research": _OkAgent()}
+        res = orch.run_daily(date="20260601")
+    assert "risk" in res and isinstance(res["risk"], dict) and "error" in res["risk"]
+    assert "selection" in res and isinstance(res["selection"], dict) and "error" in res["selection"]
+    # 失败不中断：report/sentiment 仍成功
+    assert "report" in res and "error" not in res["report"]
+    assert isinstance(res.get("sentiment"), pd.DataFrame)
+
+
+def test_orchestrator_total_steps_is_5():
+    assert TOTAL_STEPS == 5