wlr_scene_buffers need an additional wlr_scene_buffer_set_source_box()
to be scaled correctly by wlroots based on the output scale.
With this commit the buffers are now scaled correctly but, especially
on higher scale factors, may be a bit blurry.
In the future we likely want a dynamic font label generator based
on the maximal scale of all the outputs the specific buffer is currently on.
Some inspiration for that could be
https://github.com/Nefsen402/sway/blob/scene-graph/sway/sway_text_buffer.cFixes#348
