Path dependency. Insect eyes and vertebrate eyes presumably each occupy their own unique minima in the energy landscape. Even if we could describe a cost function that picked one of them as more globally optimal, moving from one to another would be hideously expensive.

It's also possible that compound eyes have an advantage at smaller eye sizes---or maybe a better way to put it: their disadvantage relative to vertebrate eyes is less pronounced for equally small eyes. Pure speculation on my part though.