When we generate bucket-level predicates on time series collections, we want to be able to generate an alwaysTrue or alwaysFalse predicate if we are comparing to an extended range date but we know that there are no extended range dates in the collection.

In order to decide when a predicate operand is "extended range" we should be taking into account the bucket rounding and span, updating this constant.

We should instead base our calculation on this value, but with rounding and span applied.