Generalize drop_nulls fusion

question: How general can we apply this to other pylibcudf aggregations that accept `null_policy` (e.g count, rank, distinct_count),

https://github.com/rapidsai/cudf/pull/22014 fuses `drop_nulls` and `n_unique` into a single pylibcudf call because `n_unique` supports null handling natively. We may be able to apply the same optimization to other operators where pylibcudf natively supports `n_unique` as well.