This pull request introduces enhancements and refactoring to the ARIMA-based time series generation and improves the organization of visualization utilities. The most significant changes include improved error handling and normalization options in the ARIMA simulator, the migration of the Shapiro-Wilk residual normality test to a dedicated visualization module, and updates to documentation and package imports for clarity and usability.

ARIMA Simulator Improvements:

• Added reversible normalization (revin parameter) to the ARIMASimulator, allowing generated series to be normalized and later restored to original scale. [1] [2] [3] [4]
• Improved input validation in ARIMASimulator by checking for NaN values and effectively zero variance in input series, with clearer error messages. [1] [2]
• Enhanced error handling: residuals that fail the white noise test now raise a ValueError instead of printing a warning, making failures more explicit.
• Added properties for accessing ARIMA model parameter names and values (param_names, params, param_items).

Visualization Refactoring:

• Migrated the plot_shapiro_wilk function from s2generator/utils/_tools.py to s2generator/utils/visualization.py and updated imports and __all__ accordingly. [1] [2] [3] [4] [5] [6] [7] [8] [9]

Documentation Updates:

• Updated README.md to highlight the new ARIMA-based time series generation method and clarify dependencies and installation instructions.

Minor Cleanups:

• Removed outdated comments and unnecessary imports, and clarified docstrings for better maintainability. [1] [2]