I'm also in the "if we're going to do this, it should probably be a new function" camp. e.g. `some1`. I'm kinda neutral on adding it at all at this point, but if we do it, it should probably be done along these lines.
As for using a semigroup based implementation, its performance is, sadly, rather awful for constructing `NonEmpty` lists, and more damning, you have no law relating the 'pure' or whatever you're using to create values and the Semigroup you're using to glue them together, so it only works for NonEmpty on a purely ad-hoc type-by-type basis. That signature isn't one we want as evidenced by the fact that if you pick m = Maybe you sure as heck won't get anything like what you expect! This is the pointed problem all over again. I'm very strongly -1 on that variant of the proposal.