No, the scaling is definitely done by the applications. I don't know how they do it, but it works.
I don't think it is possible to avoid having lines "in between" pixels when using fractional scaling. The physical pixel grid simply does not always align with the one at "100%" scale.
I specifically bought the 4K monitor to get crisp text rendering. Upscaling from 1080p (e.g. what XWayland does) is very noticeable to me. I don't see any difference between native Wayland fractional scaling, and setting Gnome to native resolution (except for the tiny UI elements).
By native scaling I mean absolutely no bitmap scaling. If rendering happened at target scale then all pixels would match grid and no artifacts would be possible. I can't even imagine how ClearType could work with bitmap scaling when it targets individual subpixels.
I think people have wildly different sensitivity to resolution. I know some that can't tell difference between 4k and 1080p and 1080p scaled on 4k screen. Some crazy people even think MacOS has good font rendering.