In order to support larger datasets it would be useful to have `nnz_t` type instead of `int` Hopefully, this will help solve https://github.com/rapidsai/cuml/issues/6910.