[FEA] Experiment try to unify implementation of non-nested types and nested types in `cudf::contains`

Currently, `cudf::contains` implements two separate code paths for non-nested types and nested types. That requires more effort to maintain all of them, comparing to just have one code path that can handle both.

Recently, there was an attempt to use just one code path to handle all data type (ref: https://github.com/rapidsai/cudf/pull/10770#discussion_r880980883) in groupby key hashing, which reported no performance regression on basic data types. As such, we should also test and try to unify the two code paths of `cudf::contains`.

Having just one code path that handles everything, maintenance will be much easier.