`make_predictions` uses faulty `test_time_value` when using `epix_slide`

[brookslogan]

(make_predictions, arx_forecaster, epix_slide) don't work together quite correctly:

make_predictions takes a given ref_time_value (= target date), fetches the data $as_of that target date, but then doesn't pass that target date through to make_predictions; instead, make_predictions assumes that the test_time_value is the latest time_value with data available, which is rarely the case: we typically work off of data with some nonzero latency (or maybe, off of some point forecasts with negative latency). So we end up with forecasts for ref_time_values + ahead - minimum_latency_among_signals, but labeled as forecasts for ref_time_values + ahead.

epix_slide might need updated to pass along the ref time value if it doesn't allow that already.

[dajmcdon]

Marking this wontfix but leaving open. The intention is to deprecate make_predictions() and rewrite the forecaster using the recipe framework. We can revisit at that time.

[dajmcdon]

Perhaps addressed by #55 .