lisa.datautils: Add df_combine_duplicates(prune=True)

When prune=False, do not remove the rows of the duplicate group, but
replace the values in output_col with the series returned by the
callback instead.