alternate ensembling option in ts_fit_forecast

[vpnagraj]

as of the v2.0.2 release (https://github.com/signaturescience/fiphde/releases/tag/v2.0.2) we now allow NNETAR to be included in the ensemble.

however, as implemented in fable the NNETAR model does not generate forecasts that are normally distributed.

per the fabletools documentation the combination modeling can only create distributional forecasts for models that generate normally distributed forecastes:

Currently distributional forecasts are only supported for models producing normally distributed forecasts.

https://fabletools.tidyverts.org/reference/combination_model.html

to get quantiles, we may want to consider an alternative method for ensembling that does not use the forecasts created from the combination modeling.