feat: 用户行为信誉分系统

- User 新增 credit_score 字段（0-200，默认100）
- 信誉分影响检测阈值系数：高分降低敏感度，低分提高敏感度
- 发布成功+1分，被拦截-2分；申诉通过+10分，驳回-5分
- 新增手动调整和批量重算信誉分接口
- admin-users 页面显示信誉分进度条，支持编辑调整

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/app/routes/spam_routes.py

@@ -32,6 +32,15 @@ def _threshold() -> float:
     return float(row.spam_threshold) if row else 0.75
 
 
+def _adjusted_threshold(user_credit: int = 100) -> float:
+    """根据用户信誉分调整阈值"""
+    base_threshold = _threshold()
+    # 系数范围：0.85 - 1.15
+    credit_factor = 1.0 + (user_credit - 100) * 0.0015
+    credit_factor = max(0.85, min(1.15, credit_factor))
+    return base_threshold * credit_factor
+
+
 @spam_bp.post("/predict")
 @jwt_required()
 def predict_one():
@@ -46,7 +55,7 @@ def predict_one():
 
     clf = _ensure_ready()
     result = clf.predict(text)
-    threshold = _threshold()
+    threshold = _adjusted_threshold(user.credit_score or 100)
     blocked = float(result["spam_probability"]) >= threshold
 
     row = SpamPredictionLog(
@@ -82,7 +91,7 @@ def predict_batch():
     clf = _ensure_ready()
     rows = []
     results = []
-    threshold = _threshold()
+    threshold = _adjusted_threshold(user.credit_score or 100)
 
     for text in items:
         content = (text or "").strip()