[FEA] row_number should work with all types of order-by columns

[mythrocks]

This is a follow-up to #12178.

In its current form, row_number has similar restrictions to the other ranking functions (rank, dense_rank, etc.), in terms of the permitted data types for the order-by column.

It would be good to remove the restrictions for row_number. While rank(), dense_rank(), and percent_rank() need to compare the values of the order-by column rows to determine the rank value, row_number() does not.