I don't know how well this technique for removing surface crud will work on other types of prints or pictures with other subjects; I'll have to do some experiments. The night scene seems to have helped the neural network "decide" that most of the specs it found on the dark background didn't belong there, so when it rebuilt the image it eliminated most of them, and the matte paper I used to make the print may also have had an effect. In any event, Photoshop's Dust & Scratches filter easily dispatched the remaining imperfections and I didn't need to do any manual spotting.
I've tried using Super Resolution on scans of 35mm negatives, but it seems mostly to have treated the surface imperfections as image detail and enlarged them. Of course, those scans were made at much higher resolution, but I think a tranmissive surface also tends to make spots more apparent than a reflective one.
As I've mentioned in another section of this forum, I'm not aware of any technical impediment to using machine learning to allow a neural network to accurately identify and eliminate surface imperfections on any type of image without degrading sharpness. Possibly the same software could be used to deal more intelligently with digital noise than current methods. It isn't clear to me whether the economic inducement is there to devote the resources to producing a commercial product, but hopefully somebody is working on it—at least as a proof-of-concept.