Both this version and the old rational version have the mouse-over issue, but I didn't look into it much.
It turns out so long as virt-to-raw and dpi-to-raw-dpi scaling is always up-scaling, which I think will always be true, the test failures (and the Dungeons & Dragons mouse-look issue) are resolved by using float intermediate values when mapping coordinates. This avoids internal rounding errors occurring during multiple scalings in `map_monitor_rect()`. I raised !9183 for it and will close this.