The very complex task you asking for, Ross. Cool video, I like to watch your videos about your ideas.
I think it will be a great idea for games you can not access the data, like yea, games on encrypted servers. Fear that with all this "google stadia" stuff there might be even more of them. Something is better than nothing, you are right.
However in general game worlds and game looks and feels is very complex thing, as you mentioned in the video.
World data, external meshes, palettes, shaders, collision. The first thing I had in mind is my internal video game purist screaming inside my head. I think Daggerfall unity was an awesome example. I like that this port exists, that Daggerfall got back its modding community, however, I love the original engine, Bethesda's Xngine. How it feels, how it renders things, and how a lot of things worked in the original game. And this port on unity is not... very accurate. Using default unity stuff, it making lower resolutions and original palette only as some post-shader "emulations". Walking physics, rain and snow effects, headbobbing effects were different, and just some unity stuff that looks weird. Speaking of drawing distance it also has limitations, and more than that, terrain generation is inaccurate. It just using the original height map of the game, but all magic done now by unity instruments, so you have now this smooth terrain with huge view distance that looks not much like the original. This game feels like a "unity game with Daggerfall assets" and I don't like this feeling. Sad that XLengine project died, I hope Lucius will come back for it after he will end with "The Force Engine", my favorite kind of projects of new engines is ones that reverse-engineer original as closely as possible, with new features added only something on top on the accurate representation of the original game.
And problem of such projects not lack of data usually, everything cam be reverse-engenereed, just lack of enthusiasm or interest in some games. There are huge amount of games that will benefited from ports, patches, improvments but there are no one really interested to doing something complex with them.
So yea, when we take the idea of "ripping off worlds, assets, games, using our or machine's perception" it will end in a very inaccurate approximation even if the software is kind of perfect. It it something that will be better used for capturing real world, thank god physics laws generally the same, than for worlds created by man. For games (at least, to which files we have acces to) this dream software for me is instruments like Ghidra, that allow to reverse compiled product back into code.
However, speaking about your ideas about "video game musem" this is absoluetly awesome idea! You know, long time ago when I played Garry's Mod I had wierd idea what if porting or making model for everything, port here all props, models, maps from a lot of games, and more, universes! This something what can be started before dream tool you are discribing, but might leed to legal issues. However for most older games I don't think anyone would care lol.
Addition: this probably will go off topic but I wondered about the game that has all the cars ever made. Or, at least, a lot and updating them. I'm not a super racing gaming fan, in my dream game I want just to go to the garage, spawn a car from a database, for example, a 1986 Opel Kadett E sedan with red color and 1.4 engine. Or, 1984 Ford ltd lx, or some 1909 De Dion-Bouton Torpedo. Be able to disassemble it in VR, assemble it back, ride it around with real characteristics, crash and repair. Wierd idea that haunts me for years, really.