feat: ENEL time-series forecasting pipeline with walk-forward validation

[Copilot]

Adds a reproducible ML forecasting pipeline for the ENEL daily stock dataset, comparing multiple models under strict walk-forward validation (no leakage).

Pipeline (enel/enel_forecast.py)

• Data loading: locale-aware parsing (Italian format: 1.234,56 → 1234.56), date sorting
• Feature engineering: Close/return lags 1–10, volume lags 1–5, rolling mean/std (windows 5/10/20), existing technicals (MACD, RSI, MAs, etc.); all features at t predict t+1
• Targets: next-day Close price (regression) and next-day direction Up/Down (classification)
• Validation: TimeSeriesSplit walk-forward (default 5 folds); scaler fit per fold to prevent leakage
• Models: Naive baseline, ARIMA(2,1,2), RandomForest, LightGBM — both regression and classification
• Outputs: prints best-model recommendation to console; saves full comparison to reports/enel_model_comparison.csv

pip install -r enel/requirements_enel.txt
# place dataset at enel/ENEL.csv then:
python enel/enel_forecast.py [--data PATH] [--splits N] [--skip-arima]

ARIMA is auto-skipped for datasets > 1000 rows (too slow for per-step re-fitting); --skip-arima forces skip.

Other changes

• enel/requirements_enel.txt: pinned deps (pandas, scikit-learn, lightgbm, statsmodels)
• .gitignore: excludes enel/ENEL.csv (user-provided, not committed)
• README.md: added ENEL forecasting section with dataset format, assumptions, and usage