Probably because to date, all raw processing software (at least that I know of) only sees Bayer arrays, and that would be Canon's default output. Even now, with being able to save dual pixel raw, Canon only uses that info so you can do AF micro adjustments after the fact, not actually generate a file with more resolution from the two sub-pixels.
Going to quad pixel AF still allows very easy bayer output, and if done, does allow very easy spatial resolution bumps by not combining the sub pixels, but does significantly bump up the post processing requirements. I suspect part of the reason why canon went to the CR3 file format over the CR2 format is to make it easier to store non-standard pixel arrays. You can save dual pixel raw files in CR2 files (like the 5DIV does), but it basically stores it as two bayer array images in sub-chunks. The CR3 format stores each color discretely in it's own chunk and the raw processor has to then read each color chunk, then combine it into a bayer array, then demosaic it. The CR3 format is a pretty big deviation to how Canon stores its sensor data over the CR2 format.
I also suspect Canon has very good reason to go quad pixel AF because it allows them to have more than 2 output gains. This is how they were able to get the DR increases and noise improvements in recent dual pixel bodies. Each sub-pixel is actually 1 stop different than the other one. The way they store it in CR2 files, again, is less than ideal, as they store the first bayer array as they normally would with both sub pixels combined, and the second bayer array with just the output of the second sub-pixel. With a quad pixel array, they'd have pretty good reason to store each sub-pixel by itself if saving quad-pixel files as it would mean a lot more flexibility when generating a full color image. That, and they could have quad gain structure where each sub pixel had 1 stop more gain than the next, giving a combined 4 stop spread between the sub pixels with which to generate an image from. This would be how they could get to 15.5+ stops (if outputting a ~24MP bayer array where all the sub pixels are combined). They could keep a 12 or 14 bit AD, and have 4 gain outputs. If they stored each sub pixel separately, they don't even have to store 16 bits per pixel, they could still do 12 or 14 bits, then when generating the full color image after the fact in their DPP software, store the full resulting RGB as a 16 bit TIFF file. I wouldn't be surprised in the least if it was just a straight 12 bit ADC (for speed), and they just use the multi-gain to get the DR.