Yes, two different issues here :
Lenses that are allowed to project their image on a sperical surface can have an easier time if designed to do so. Here the intermediary matters.
If the final output is organized so as to be printed on the inside of a goldfish bowl or on slightly spherical paper for slightly spherical picture frames, no problem in going from a surface sensor forming part of a sphere to the desired final result.
Projecting anyhing curved, spherical or simply a plane (at an angle) onto a flat surface entails distortion. Mercator is just one (very ancient) way of projecting a sphere onto a flat map. Any correction will need to add, "enlarge" or delete, "diminish" pixels. Expensive lens-by-lens correction will be replaced by a cheap once-and-for all program for all instances of that lens\sensor combo.
For a small solid angle piece of a sphere, the distortions will be correspondingly small.
A suitably crinkly and continuously malleable surface sensor might even correct for "moustache distortion" and adapt to lenses that change curvature of field with how their controls are set.