Implementing this kind of line-by-line interface efficiently is only possible if one has word boundary information (actually, the line boundary) from the OCR. That kind of information was not available when DP started and to add it into the current DP would be so much effort as to make it out of the question.
I've been thinking about the possibility of this kind of interleaved bitmap/txt at least as an option for SR? You can presumably flag the linebreaks inside the png in a non-intrusive way. It would require generating SR in a file format that can mix bitmap and txt, for example either html or rtf. If the linebreaks are coded non-intrusively inside the png then it is back-compatible with what you have now. Then you just need separate optional page editor software, and you have given the proofers the option of doing it line-by-line.