No, the perfect pixel is a sinc() function of appropriate bandwidth. A pixel of the form "dot with a dark area around it" is a better approximation of sinc() than pixel of the from "fill all the available space evenly".

And it doesn't mean the bright part of a pixel should be as small as possible, no, it should be tuned for bandwidth.

Even eye receptors (cones) approximate sinc() in the same fashion.