Can we add more types to libcudf?

[GregoryKimball]

We did an experiment to add more fixed width types to cuDF, so evaluate the impact on binary size and performance. The experiment stressed dispatcher growth while keeping storage consistently 64-bit.

Binary size increased linearly at about 2% per type Even adding one more type brings strong sorting regressions. This is likely to block the binary type addition project (see #23658)

[GregoryKimball]

Some items in flight will improve the sorting cliff with additional types:

• Replace hash groupby internals with HashCSR #24050 will have follow-on work to extend hash CSR data structure to sort-based groupby, avoiding the explicit sort.
• Use primitive lexicographic comparison for numeric sorting #24138 will allow primitive types to use a more efficient row operator, even for multi-column sorts.
• We expect that multi-column string sorting could still hit the performance cliff