sub-logical-pixel slivers of the background are visible at edges of scaled outputs

When running on a scaled monitor without gaps, the background potentially shows through at the bottom and right edge of the screen, even with windows open nominally taking up 100% of the space. The underlying cause is rounding, in two places:

1. The requested window size is necessarily an integer number of logical pixels, so it's floored from the desired tile size (request_tile_size). Unfortunately, there's nothing we can do about this directly, but see below.
2. compute_working_area is rounded to an integer number of logical pixels, since it comes from Smithay's non_exclusive_zone.

The main problem is #⁠1, which seems to be a persistent source of issues (e.g. #1402). However, #⁠2 makes it worse, since it means we're rounding twice, potentially leaving a sub-logical-pixel gap before computing the desired tile size and rounding again. This at least can be fixed straightforwardly by adjusting the working area to meet the output edge whenever it's within 1 logical pixel of it.

For #⁠1 there's no perfect fix. My preference would be that, if gaps are off, any such slivers should be occupied by a sub-logical-pixel thickening of the window border. This is what I intuitively expect to happen, since usually with a tiling WM one can cover the background entirely. The downside is that the border width can then vary slightly between edges. This also doesn't do anything about the situation when borders are disabled (though in that case it really seems like there's nothing we can do).

I can write a PR for these changes if they're deemed suitable. I figure the fix for #⁠2 should be uncontroversial, since it's a simple change within one function and just fixes a double-rounding bug. The fix for #⁠1 is more involved, as it requires tiles to keep track of their requested size and requires the layout engine to use that requested tile size for windows which conform to their requested window size. It's also a genuine change to the rendering logic rather than a simple bugfix. (On the plus side, it should also fix #1402).

prefer-no-csd
output "HDMI-1" {
    scale 1.75
}
layout {
    gaps 0
    border {
        width 2
    }
}

System Information

• niri version: 25.11 (Nixpkgs)

• Distro: NixOS

• GPU: Intel iGPU

• CPU: Intel Core i5-6600K