Forward incidence_period to forecaster in get_predictions and/or remove default in baseline_forecaster

[brookslogan]

In evalcast:

• get_predictions takes argument incidence_period and uses it to define the evaluation data, but does not appear to pass it along to the forecaster (anymore)
• baseline_forecaster (as well as cmu-delphi/covid-19-forecast's animalia framework) accepts incidence_period as an argument, but defaults to "epiweek"

The combination of the two points above makes baseline_forecaster output weekly predictions even when daily predictions seem to be requested. (Predictions are for epiweeks, evaluation is based on day-level data, errors are large.)

The intended approach in the current framework is likely the following:

get_predictions(baseline_forecaster,
                incidence_period = "day", <other args>,
                forecaster_args = list(incidence_period="day", <other forecaster args>))

Some potential resolutions:

• forward incidence_period to the forecaster in get_predictions; this prevents repetition of the incidence_period argument, but requires all forecasters to accept and check an incidence_period argument
• remove the default behavior for incidence_period in baseline_forecaster (and do the same for animalia forecasters); an error should be generated when attempting to run these forecasters without specifying incidence_period in the forecaster_args; this requires specifying the incidence_period twice, though (side note: the "reverse" case of specifying incidence_period within forecaster_args but not directly to get_predictions does eventually produce an error, after appearing to download a mix of weekly and daily data, at least for even numbers of requested predictions --- perhaps the default behavior in get_predictions also needs alteration)
• both of the above