The indexes passed to paragraph.getBoxesForRange are based on UTF-16 code
units.  If the caret is at a character that is encoded as multiple code units,
then the start and end indexes should be adjusted accordingly.