I mean fullscreen within the VM's desktop (no need for normal GNOME/KDE/whatever desktop), which itself may not be fullscreen on host OS. It would act like a native app. If you quit the browser it shuts down the VM.
I would expect a generic resolution like 1920x1080 to convey much less identifiable information that some random 1583x1176 that the user might resize tor browser window to.
The idea is to not change the window size at all from the default. If this advice is followed, you minimize the possible information leak. In your example, 1583x1176 tells us that your system is capable of rendering at least that size. Given the unusual numbers, we further suspect you're not maximized on a system capable of an 1176px tall browser (much fewer of those than 1920x1080). While not uniquely identifying, it's a piece of the puzzle.
A better idea would be to simply not leak any of this information at all, or if it must, return some generic 1080p regardless of the actual size. It's a terrible UX to restrict yourself to the default window size (and depending on the window manager, the default window size might not even be respected). Plus, it's so easy to accidentally change the browser window size.
It is basically impossible to fully anonymize a browser as long as JavaScript+plugins are running. EFF's Panopticlick [1] and browserleaks [2] are good at explaining some of these fingerprinting vectors.
This is worst part of Tor browsing for me. 1000x800?! Are you kidding! Must be upped to macbook's 1680, or whatever is most popular resolution of a modern display