The two "half-pixels" share the same front-lens so they probe the same spatial info and therefore are not increasing the spatial resolution. In order to double e.g. the horizonal resolution they would have to double the little frontlenses too but then they could not use it in the same way for phase contrast autofocus...
That makes sense.
Moving on, can you tell me about these little lenses? Are there literally 20million little tiny lenses?
Yes, there are.
And as already noted, these 20 million tiny lenses determine the image resolution - even though there's a 40mp sensor underneath.
The thing is, this 40mp sensor would have likely had quite poor image quality if it wasn't doing the 2->1 merging of signals.