feat: 添加可配置场景注册表与规则引擎

- 新增场景基类、配置加载、注册表与原语模块
- 添加 r10_refund_split 规则及场景 JSON Schema
- 扩展 scan 引擎与 scenarios API
- 新增场景注册表/配置/集成测试
- 更新前端 App、api、labels 支持新场景

backend/app/scenarios/churn_fraud.py

@@ -8,6 +8,9 @@ from __future__ import annotations
 
 from dataclasses import dataclass
 
+from app.scenarios.base import BaseScenario, ClueDraft, ScanOutcome
+from app.scenarios.registry import register_scenario
+
 
 @dataclass
 class CohortPoint:
@@ -83,3 +86,50 @@ def build_rationale(finding: ChurnFinding, mismatch: float) -> str:
     return (
         f"未见明显断崖退订，但佣金与业务质量不匹配度为 {mismatch:.0%}，建议关注。"
     )
+
+
+@register_scenario
+class ChurnFraudScenario(BaseScenario):
+    """场景二养卡骗补检测的插件封装：时序断崖 + 佣金质量不匹配 → 线索草稿。"""
+
+    code = "R9"
+    title = "疑似养卡骗补（脉冲增长+规律退订）"
+    risk_domain = "成本"
+    label = "养卡骗补"
+    score_threshold = 0.5  # 与原 run_churn_scan 的 score>=0.5 一致
+
+    def __init__(
+        self,
+        retention_curve: list[CohortPoint],
+        commission_paid: float,
+        active_ratio: float,
+        zero_usage_ratio: float,
+        channel_key: str,
+    ) -> None:
+        self.retention_curve = retention_curve
+        self.commission_paid = commission_paid
+        self.active_ratio = active_ratio
+        self.zero_usage_ratio = zero_usage_ratio
+        self.channel_key = channel_key
+
+    def scan(self, session, *, data_version_id=None) -> ScanOutcome:
+        finding = detect_pulse_decay(self.retention_curve)
+        mismatch = commission_quality_mismatch(
+            self.commission_paid, self.active_ratio, self.zero_usage_ratio
+        )
+        score = churn_risk_score(finding, mismatch)
+        draft = ClueDraft(
+            score=score,
+            rationale=build_rationale(finding, mismatch),
+            evidence={
+                "cliff_month": finding.cliff_month,
+                "max_drop": finding.max_drop,
+                "commission_paid": self.commission_paid,
+                "active_ratio": self.active_ratio,
+                "zero_usage_ratio": self.zero_usage_ratio,
+                "mismatch": mismatch,
+            },
+            subjects={"channel": self.channel_key},
+            amount_involved=self.commission_paid,
+        )
+        return ScanOutcome(scanned_count=len(self.retention_curve), drafts=[draft])