Backtesting and evaluation

Toolset tsforecast_backtest: 2 tools.

Tool	What it does	Notes
`tsforecast_cross_validation`	Rolling-origin backtest of one series (y) or a panel (data): for each of n_windows origins (step_size apart) the models are fitted on the history before it (all of it, or the last input_size points; re-estimated every refit windows) and forecast the next h periods, which are scored against the actuals.
`tsforecast_evaluate`	Scores model forecasts against actuals with any of 25 metrics: point (MAE, MSE, RMSE, MAPE, sMAPE, WAPE, ND, bias, cumulative error, periods in stock, relative MAE, Tweedie deviance, LINEX), scaled by the in-sample seasonal naive (MASE, MSSE, RMSSE, scaled PIS), quantile (pinball, multi-quantile, scaled CRPS) and interval (coverage, Winkler, calibration).

Inputs, limits and outputs are described in Fincept TS Forecast. Full schemas: fincept_describe_tool.