[FEA] Implement lists::drop_list_duplicates using cudf::distinct

[ttnghia]

I just realized that lists::drop_list_duplicates can be implemented using cudf::distinct by using the following approach:

1. Peel out child column of the input lists column
2. Generate labels for the child elements
3. Add the labels column into the keys table that input into cudf::distinct
4. The output labels of unique elements need to be grouped and maintain their group order, this can be achieved easily by changing cudf::distinct to use a gather map generated from a sequence of linear indices.
5. The labels output from cudf::distinct are used to reconstruct the final lists column without duplicates.

Such approach can be implemented if we have cudf::distinct supports:

• duplicate_keep_option parameter ([FEA] Support duplicate_keep_option in places that use unordered map #11050), and
• nan_equality parameter, and
• Step 4 above.

The only concern here is how to enforce step 4 above. Adding parameters is easy, even that is breaking change. Guarantee some internal implementation is more difficult, as devs may accidentally + silently change it. Maybe explicitly writing this into the public interface doxygen and adding unit tests can help.

[ttnghia]

Using this approach, my local implementation has only about 50 LOC, instead of >600 LOC as in the current implementation.

[sameerz]

Depends on #11149