Look into applying DECIMAL128 sum aggregation overflow checking for window aggregations

[jlowe]

#4688 adds a new technique for detecting overflows in DECIMAL128 sum aggregations, and we should investigate whether it makes sense to do something similar for window sum aggregations. Unlike normal sum aggregations, window sum aggregations are on sorted data ,so the use of sort-based aggregations shouldn't be a performance concern when we hint libcudf about the sorted input. However we are still performing special calculations for overflow checking that is different than what we're doing for sum aggregations, and it may make sense to commonize this code.