Are you intending to buy any native lenses, or are you buying the body strictly for use with adapted lenses?
If you don't intend to buy native lenses, you're free to choose any camera body out there. In this case, you're not locking yourself to a particular lens mount and are free to change systems later on if future generations of bodies have different strengths. In this case, unless you particularly need dual card slots, I would go for the Z7 - it has marginally more resolution than the A7r3, can shoot at ISO 64 (presumably, you're shooting things that don't move much, since you're talking about adapting manual focus lenses), has a shorter flange distance, potentially allowing a slightly greater range of lenses (although the number of lenses that can fit on a Z-mount but not on an E-mount must be miniscule, outside of native Z-mount lenses) and better weather sealing. Even if Nikon goes belly-up in the future, you're not risking anything, because you won't actually be buying anything (other than batteries and other accessories) which are Z-mount-dependent - if, for instance, Fuji or Olympus later release a better body, you could just as easily adapt your lenses to that new body.
If you do intend to buy native lenses, you also need to consider what lenses are available, or will likely become available in the near future, for each lens mount, as well as the likely future direction of cameras using that lens mount. In this case, I would go for the A7r3. It is a much more mature system, with greater OEM and third-party support, has much better AF (which comes into play with native lenses) and a much more certain future.
As for pixel shift, what are you shooting?
Unless you're shooting things that don't move at all, Panasonic pixel shift is near-useless. It moves the sensor by half a pixel in each direction, combining four images to create a final image with double the linear resolution of the sensor (or four times the pixel count). Whenever movement is detected, a single frame is interpolated to the higher final resolution, creating an area which has visibly lower resolution (and, sometimes, aliasing artifacts) compared to the surrounding area. This can appear quite offputting - when looking at a pixel-shift image, what you see is some areas which are extremely detailed, but patches of movement-containing area which look almost smeared by comparison, since they are interpolated from a single frame. Sony pixel shift is marginally more useful - it doesn't increase the final resolution, but, rather, measures true RGB numbers for each pixel, rather than applying Bayer interpolation. Movement-affected areas can then be masked over by a single, Bayer-interpolated frame from the four source images. This is less offputting than masking applied to half-pixel-shift images, since the masked areas contain no less spatial resolution than the areas using all four source images; only colour detail is affected.
'Things that don't move at all' excludes most landscapes. Water ripples, leaves and grass move slightly (even one pixel of movement is enough) and, in urban environments, lighting from headlights and artificial light sources change between frames, even if buildings don't themselves move and there is no traffic visible within the frame. What it includes, and what pixel shift is mostly useful for, is things like in-studio still life and product photography, artwork reproduction and microscopy of dead things on slides. Unless you are primarily doing one of these things, I wouldn't factor pixel shift into your choice of camera at all.
