After some more reading, trial and error (temp hosting):
This is the dark side of a planet near the dawn line; one can see a bit of the lit part in the bottom right corner. The bright dot on the left is a star, the other isolated points are space dust.
I targeted the coastlines to make tests more easy, so it looks a bit ugly and silly (actually on this picture the criteria was still including water bodies). As far as I understand it, I indeed flipped the depth coordinate of the normal vectors (Z in [0..128] instead of [128..255]).
The downside is that the brighter the point is on the dark side, the darker it will appear on the lit side. So it looks terrible especially on the coastlines which are normally pale, but I think it's still usable for cities which can be dark patches. Maybe it will be possible to mitigate this by picking a bright color for them.