Act as a Lead Data Scientist. Provide a comprehensive technical tutorial on forecasting multivariate time-series data for retail sales. Assume the dataset includes daily sales figures, holidays, price promotions, and weather data. The guide must cover: 1) Data preprocessing steps including stationarity tests (ADF/KPSS) and seasonality decomposition (STL), 2) Feature engineering for lagged variables and rolling window statistics, 3) Comparing the performance of SARIMAX, Prophet, and LSTM models, 4) Implementing cross-validation for time-series (TimeSeriesSplit) to prevent look-ahead bias, and 5) Evaluation metrics beyond RMSE such as MASE and scaled error. Provide Python code snippets using pandas, statsmodels, and sklearn.