Set a max window size for training data

It may happen that older data negatively affect forecasts. For example, the impact of seasonality can change over time.

Would it be possible to add a parameter to set a maximum window size for the data considered (eg. max 2 years in the past)?

In particular, for `cross_validation`, we would have a maximum window size for the historical data, so we can discard older data in cross-validation (instead of having a forever-expanding window).

Thanks