feat(backtest): 成本模型拆分为佣金+印花税(仅卖出)+滑点 (#46)

* feat(backtest): 成本模型拆分为佣金+印花税(仅卖出)+滑点

MatcherConfig 新增 commission_pct/stamp_tax_pct 及 buy_cost_pct()/
sell_cost_pct() 方法, 撮合三处现场统一改用方法计算。

印花税单边(仅卖出)是 A 股与美股的本质差异: 旧的 fees_pct 双边模型
把印花税摊到买卖两腿, 会高估买入成本、低估卖出成本。拆分后买入腿=
佣金+滑点, 卖出腿=佣金+印花税+滑点。

commission_pct 未设时回退 fees_pct, 印花税未设时为 0, 完全向后兼容,
既有 12 个 portfolio 测试不变。strategy/api(含 SSE + 缓存键)全链透传。

新增 tests/backtest/test_cost_model.py 覆盖向后兼容/拆分/印花税仅卖出/
优先级/撮合传导 5 个用例。

* fix(backtest): 打通成本模型前端链路并修复 cancel 缓存键

子代理审查发现 PR3 半成品缺陷, 本次补全:

C1 (正确性): /strategy/cancel 的 _make_job_key 漏传 commission_pct/
stamp_tax_pct, 一旦用户设成本参数, cancel 算出的 job_key 与 stream 失配
导致取消静默失败。补 None-aware 解析对齐两侧口径。

C2 (完整性): 前端 SSE/sync 链未透传新字段, 新成本模型从 UI 够不到。
- backtestTask.ts / api.ts: 类型 + query 透传 commission_pct/stamp_tax_pct
- StrategyBacktest.tsx: 佣金映射到 commission_pct, 新增印花税(千分之)输入,
  默认 1 (A股千1), 映射 stamp_tax_pct; 状态持久化
- storage.ts: stampTax 类型声明

测试加固:
- test_stamp_tax_only_deducts_on_sell_leg: 去掉硬编码 shares=9900,
  改为断言两次运行 shares 相等 (守护买入腿 sizing 不受卖出成本污染)
  + 从结果反推卖出市值校验差额
- 新增 commission_pct=0.0 非 None 边界 (防 falsy 回退)
- 新增 job_key 区分成本参数的回归测试 (守护 C1)

后端 26 测试全绿; 前端 tsc 仅剩既有 baseUrl deprecation (非本次引入)。

backend/app/api/backtest.py

@@ -191,6 +191,8 @@ class StrategyBacktestRequest(BaseModel):
     entry_fill: Literal["close_t", "open_t+1"] | None = None
     exit_fill: Literal["close_t", "open_t+1"] | None = None
     fees_pct: float = 0.0002
+    commission_pct: float | None = None
+    stamp_tax_pct: float | None = None
     slippage_bps: float = 5.0
     max_positions: int = 10
     max_exposure_pct: float = 1.0
@@ -224,6 +226,8 @@ def strategy_run(req: StrategyBacktestRequest, request: Request):
         entry_fill=req.entry_fill,
         exit_fill=req.exit_fill,
         fees_pct=req.fees_pct,
+        commission_pct=req.commission_pct,
+        stamp_tax_pct=req.stamp_tax_pct,
         slippage_bps=req.slippage_bps,
         max_positions=req.max_positions,
         max_exposure_pct=req.max_exposure_pct,
@@ -277,8 +281,9 @@ def _make_job_key(
     max_positions: int, max_exposure_pct: float, initial_capital: float, position_sizing: str,
     params: str | None, overrides: str | None,
     mode: str = "position", holding_days: int = 5,
+    commission_pct: float | None = None, stamp_tax_pct: float | None = None,
 ) -> str:
-    raw = f"{strategy_id}|{symbols}|{start}|{end}|{matching}|{entry_fill}|{exit_fill}|{fees_pct}|{slippage_bps}|{max_positions}|{max_exposure_pct}|{initial_capital}|{position_sizing}|{params}|{overrides}|{mode}|{holding_days}"
+    raw = f"{strategy_id}|{symbols}|{start}|{end}|{matching}|{entry_fill}|{exit_fill}|{fees_pct}|{slippage_bps}|{max_positions}|{max_exposure_pct}|{initial_capital}|{position_sizing}|{params}|{overrides}|{mode}|{holding_days}|{commission_pct}|{stamp_tax_pct}"
     return hashlib.md5(raw.encode()).hexdigest()[:12]
 
 
@@ -293,6 +298,8 @@ async def strategy_stream(
     entry_fill: str | None = None,
     exit_fill: str | None = None,
     fees_pct: float = 0.0002,
+    commission_pct: float | None = None,
+    stamp_tax_pct: float | None = None,
     slippage_bps: float = 5.0,
     max_positions: int = 10,
     max_exposure_pct: float = 1.0,
@@ -341,6 +348,7 @@ async def strategy_stream(
         fees_pct, slippage_bps, max_positions, max_exposure_pct, initial_capital, position_sizing,
         params, overrides,
         mode, holding_days,
+        commission_pct, stamp_tax_pct,
     )
 
     _cleanup_stale_jobs()
@@ -374,6 +382,8 @@ async def strategy_stream(
                 entry_fill=entry_fill,
                 exit_fill=exit_fill,
                 fees_pct=fees_pct,
+                commission_pct=commission_pct,
+                stamp_tax_pct=stamp_tax_pct,
                 slippage_bps=slippage_bps,
                 max_positions=int(max_positions),
                 max_exposure_pct=float(max_exposure_pct),
@@ -447,6 +457,10 @@ async def strategy_cancel(request: Request):
     p = parse_qs(qs)
     def _get(key: str, default: str = "") -> str:
         return p.get(key, [default])[0]
+    def _get_opt_float(key: str) -> float | None:
+        # 可选成本参数: 缺省或空串 → None (与 stream 侧 float | None 口径一致, 保证 job_key 对齐)。
+        v = _get(key)
+        return float(v) if v else None
     job_key = _make_job_key(
         _get("strategy_id"),
         _get("symbols") or None,
@@ -465,6 +479,8 @@ async def strategy_cancel(request: Request):
         _get("overrides") or None,
         _get("mode", "position"),
         int(_get("holding_days", "5")),
+        commission_pct=_get_opt_float("commission_pct"),
+        stamp_tax_pct=_get_opt_float("stamp_tax_pct"),
     )
     job = _running_jobs.get(job_key)
     if job and not job.done: