This needs to look to the user as if he is selecting text, even though he's really just selecting an area of the rendered page that aligns with the rendered characters within the range of text being selected: i.e. we probably don't need to render the text on top of the selection, we can just draw translucent rectangles on top of the page, as we do for the text-search feature.

FPDFText_GetRect should be able to provide the geometry of the selection rectangle nearest to a start point and an end point.