Originally _normalize_multi would deduplicate the values in the cell using a set.  I changed my mind and changed this to allow duplicates.  If the original cell had dups, I want the dictionary cell to have dups.