feat: add LSTM, hybrid ensemble, PCA, scaler, ATR stops, rolling window

Major upgrade to the ML engine:
- LSTM model type: 2-layer PyTorch LSTM with early stopping, GPU support
- Hybrid mode: LSTM (60%) + XGBoost (40%) with agreement gating
- StandardScaler normalization (critical for LSTM)
- PCA dimensionality reduction (configurable variance retention)
- ATR-based dynamic stop-loss/take-profit adapting to volatility
- Rolling window retraining for more realistic time series validation
- Updated LLM system prompt with docs for all new parameters
- All backward compatible (xgboost/lightgbm/catboost still work)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

config/initial_config.json

@@ -1,43 +1,40 @@
 {
-  "model_type": "xgboost",
+  "model_type": "hybrid",
   "features": {
-    "technical_indicators": [
-      "RSI_14", "RSI_7", "RSI_21",
-      "MACD_line", "MACD_signal", "MACD_hist",
-      "BB_upper", "BB_lower", "BB_width",
-      "ATR_14",
-      "SMA_5", "SMA_10", "SMA_20", "SMA_50", "SMA_200",
-      "EMA_5", "EMA_10", "EMA_20", "EMA_50",
-      "OBV",
-      "stoch_k", "stoch_d",
-      "williams_r",
-      "CCI_20",
-      "ROC_10",
-      "keltner_upper", "keltner_lower"
-    ],
+    "technical_indicators": ["RSI_14", "RSI_7", "MACD_line", "MACD_signal", "MACD_hist", "BB_upper", "BB_lower", "BB_width", "ATR_14", "SMA_20", "SMA_50", "EMA_10", "EMA_20", "OBV", "stoch_k", "stoch_d", "williams_r", "CCI_20", "ROC_10"],
     "lookback_periods": [3, 5, 10, 20],
     "use_volume_features": true,
     "use_volatility_features": true,
-    "use_candle_patterns": true,
+    "use_candle_patterns": false,
     "use_lag_features": true,
-    "lag_periods": [1, 2, 3, 5]
+    "lag_periods": [1, 2, 3, 5],
+    "use_pca": true,
+    "pca_variance": 0.95,
+    "use_scaler": true
   },
   "target": {
     "type": "classification",
-    "direction": "long",
-    "horizon_candles": 6,
-    "threshold_pct": 1.0
+    "direction": "both",
+    "horizon_candles": 8,
+    "threshold_pct": 1.5
   },
   "hyperparameters": {
-    "learning_rate": 0.05,
-    "max_depth": 6,
-    "n_estimators": 500,
+    "learning_rate": 0.001,
+    "max_depth": 5,
+    "n_estimators": 300,
     "subsample": 0.8,
     "colsample_bytree": 0.8,
     "min_child_weight": 5,
-    "gamma": 0.1,
+    "gamma": 0.3,
     "reg_alpha": 0.1,
-    "reg_lambda": 1.0
+    "reg_lambda": 2.0,
+    "lstm_hidden_size": 128,
+    "lstm_num_layers": 2,
+    "lstm_dropout": 0.3,
+    "lstm_epochs": 100,
+    "lstm_batch_size": 64,
+    "lstm_sequence_length": 20,
+    "lstm_patience": 10
   },
   "strategy": {
     "entry_threshold": 0.60,
@@ -47,13 +44,19 @@
     "trailing_stop_pct": 1.5,
     "position_sizing": "confidence_scaled",
     "max_position_pct": 100,
-    "min_confidence_to_trade": 0.55
+    "min_confidence_to_trade": 0.55,
+    "dynamic_sl_tp": true,
+    "atr_sl_multiplier": 1.5,
+    "atr_tp_multiplier": 3.0
   },
   "training": {
     "walk_forward_windows": 5,
     "train_pct": 0.7,
     "validation_pct": 0.15,
-    "test_pct": 0.15
+    "test_pct": 0.15,
+    "rolling_window": true,
+    "rolling_train_size": 2000,
+    "rolling_test_size": 200
   },
   "timeframe": "4h"
 }