Because that would correspond 1 to 1 with the actual visible letter boxes, and the whole system of catching events outside the "real button sizes" was developed to be more forgiving than "you have to click precisely in the box".

My understanding is it's not just about including or not including some extra fixed clickable padding around each square (in which case it indeed wouldn't harm to show the whole area), but about dynamically adapting the area, based on frequent off-target clicks, probabilities, etc.