fix(ai-engine): sync FEATURE_COLS with trained models (82→102 features)

- Load feature columns dynamically from feature_cols.json
- Add 20 missing odds_*_present boolean flags to fallback list
- Fixes LightGBM 'features in data (82) != training data (102)' crash

ai-engine/models/v25_ensemble.py

@@ -141,8 +141,10 @@ class V25Predictor:
     Each market (MS, OU25, BTTS) has its own trained models.
     """
     
-    # Feature columns (82 features, NO target leakage)
+    # Feature columns — loaded dynamically from feature_cols.json to stay
-    FEATURE_COLS = [
+    # in sync with the trained models.  The hardcoded list below is only a
+    # fallback in case the JSON file is missing.
+    _FALLBACK_FEATURE_COLS = [
         # ELO Features (8)
         'home_overall_elo', 'away_overall_elo', 'elo_diff',
         'home_home_elo', 'away_away_elo',
@@ -178,6 +180,17 @@ class V25Predictor:
         'odds_ht_ou15_o', 'odds_ht_ou15_u',
         'odds_btts_y', 'odds_btts_n',
         
+        # Odds Presence Flags (20)
+        'odds_ms_h_present', 'odds_ms_d_present', 'odds_ms_a_present',
+        'odds_ht_ms_h_present', 'odds_ht_ms_d_present', 'odds_ht_ms_a_present',
+        'odds_ou05_o_present', 'odds_ou05_u_present',
+        'odds_ou15_o_present', 'odds_ou15_u_present',
+        'odds_ou25_o_present', 'odds_ou25_u_present',
+        'odds_ou35_o_present', 'odds_ou35_u_present',
+        'odds_ht_ou05_o_present', 'odds_ht_ou05_u_present',
+        'odds_ht_ou15_o_present', 'odds_ht_ou15_u_present',
+        'odds_btts_y_present', 'odds_btts_n_present',
+        
         # League Features (4)
         'home_xga', 'away_xga',
         'league_avg_goals', 'league_zero_goal_rate',
@@ -199,6 +212,24 @@ class V25Predictor:
         'home_goals_form', 'away_goals_form',
     ]
 
+    @staticmethod
+    def _load_feature_cols() -> list:
+        """Load feature columns from feature_cols.json, falling back to hardcoded list."""
+        feature_json = os.path.join(MODELS_DIR, 'feature_cols.json')
+        try:
+            if os.path.exists(feature_json):
+                with open(feature_json, 'r', encoding='utf-8') as f:
+                    cols = json.load(f)
+                if isinstance(cols, list) and len(cols) > 0:
+                    print(f"[V25] Loaded {len(cols)} feature columns from feature_cols.json")
+                    return cols
+        except Exception as e:
+            print(f"[V25] Warning: could not load feature_cols.json: {e}")
+        print(f"[V25] Using fallback feature columns ({len(V25Predictor._FALLBACK_FEATURE_COLS)} features)")
+        return V25Predictor._FALLBACK_FEATURE_COLS
+
+    FEATURE_COLS = _load_feature_cols.__func__()
+    
     # Model weights for ensemble
     DEFAULT_WEIGHTS = {
         'xgb': 0.50,