We can do it now that it's non-Copy. This also fixes a new stack
overflow when running the random test in debug mode (which somehow
occurs even though it's skipped in debug mode) that appeared after
adding LayoutPart for some unbeknownst to me reason.
I didn't properly update it for the Smithay refactor. It was reading
initial_configure_sent too early. This worked before when niri had to reset it
manually, but it no longer works now that it is automatically reset already
before entering this function.
This can happen if a surface unmaps by committing a null buffer and then
immediately does the initial commit without a sync roundtrip, while there are
pending configures from the compositor in-flight. In this case, the surface
cannot tell that the pending configures were meant for "before unmapping" and
considers them to be the new initial configure.
From our point of view, we don't get to do a proper initial configuration
sequence in this case, and receive a mapping commit without our initial
configure state. We cannot really do much about it, but it is not an error when
this specific situation happens.
This turned out to require quite a few changes.
We keep track of the tile resize animation progress separately now, in order to
provide a resizing black fullscreen backdrop for non-resizable windows.
The window is always rendered in the middle of the tile, which once again aids
with the resizing black fullscreen backdrop.
The backdrop itself will fade in from transparency so that it's less jarring.
The resize animation now keeps track of the fullscreen progress to deal with
the case where an unfullscreen resize is interrupted by another non-fullscreen
resize. In this case, the fullscreen progress continues animating to avoid
sudden disappearance of the fullscreen backdrop.
Some things like border visibility switch to this fullscreen progress once again
to avoid jarring appearance/disappearance.
The border radius animates in accordance with the fullscreen progress to match
the visuals.
