I don't think they're just taking the images and projecting them onto each lens of the vision pro. Forgetting about inter-ocular distance for a second - that just wouldn't work for head tracking and letting you move around the image.
Instead I assume they're using the pair of lenses to approximate a 3d model which allows for display at a variety of angles (all reasonably close to where the camera is), and then rendering from that model for the person looking at the image based on the angle they are actually looking at it from. Which solves the interocular distance problem, because you render for each eye based on where the persons eye is resulting in a distance between the two display images that doesn't have to match the camera.
Instead I assume they're using the pair of lenses to approximate a 3d model which allows for display at a variety of angles (all reasonably close to where the camera is), and then rendering from that model for the person looking at the image based on the angle they are actually looking at it from. Which solves the interocular distance problem, because you render for each eye based on where the persons eye is resulting in a distance between the two display images that doesn't have to match the camera.