Omit a few more days from the "other" time series model training data

[damonbayer]

Since this model does not account for right truncation, and the "other" counts are relatively stable over time, we can probably achieve better "other" forecasts by omitting a few more days from the fitting period and extending the forecast a few days into the future.