Add model bagging

[mitchelloharawild]

forecast::baggedModel()

It should accept a model definition for decomposition, and then use a time series bootstrap to produce bootstrapped residuals for the decomposition, which are then recombined to give bootstrapped data.

The resulting series are then applied to a model definition for modelling. Components including fitted values and residuals are averaged, and forecasts should produce a set of point forecast paths.

[mitchelloharawild]

Potential interface:

 bagged_model(<model_spec>, bag_decomposition(<decomposition_spec>, times = 100))

[mitchelloharawild]

Alternative interface, based on generate() to produce simulated data similar to the response:

 bagged_model(<model_spec>, <generate_spec>, ...)

Some arguments for generate() will need to be passed by ....

Ref: tidyverts/feasts#122