Text this: Side information-driven image coding for hybrid machine–human vision