This would be possible for D3D9 -- D3D11, not so much because texelFetch stuff can exist there.
(unless it was exposed as just more display modes there)
So, technically speaking, what would be the gist of it when using Vulkan?
gamescope already provides such functionality and can be used with DXVK.
Downsampling is a lot more complex than just creating higher-resolution render targets, since you also need to take into account that many applications will expect the window size to match, mouse input has to be remapped, etc.
Nothing extracted yet.
Is it possible to implement a sort of supersampling or downsampling algorithm in DXVK as found in AMD's VSR, Nvidia's DSR, GeDoSaTo, and SoTH?