🎯 Enhance directory structure and CV metrics

✨ Features:
- Add test_data_name to predict mode to prevent result overwriting
- Optimize directory structure per mode (tune/train/predict)
- Add CV mean/std for R² and MAPE in summary.json
- Enhanced plots with best fold + CV mean curves with shaded std

🗂️ Directory Structure:
- tune: only hyperparameters + logs
- train: weights + results + plots + history + metrics
- predict: results + predictions + metrics (with test_data_name suffix)

📊 Enhanced Plots:
- Best fold training curves (top subplot)
- K-fold CV mean with ±1 std shading (bottom subplot)
- Backward compatible with single training runs

🔧 Technical:
- Fixed numpy import issue in workflow_manager
- Enhanced save_results() with conditional directory creation
- Updated visualization.py for dual subplot CV plots

Author: Sorooshi