Also remove the item about one Replacement Character per invalid UTF-8
sequence, because how many bytes form a sequence?  Two?  Three?  Four?
Better just put one Replacement Character per invalid byte, like is
done now.