sql.py: Added mk_track_data_error() and use it in cast(). This also ensures that if only one source column's row in the CROSS JOIN violates a unique constraint, other source columns' rows are still inserted.