As far as I know, all these hybrid imaging+PDAF sensors have to do is divert a few hundred of the millions of photosites from image detection to AF detection, with the remaining 99.999% of sites still doing the normal job, with the normal Bayer CFA. The relatively few and tiny "holes" are then filled by interpolation from nearby "imaging sites".
I wonder how this affects third party raw conversion software? Does the raw output have the AF sensor gaps already filled by special interpolation, or does the raw software need to know where the gaps are, and handle then along with the normal demosaicing interpolation?