Use damage requests to optimize QML compositor rendering

As far as I can tell, we're rendering the entire client each time it updates, regardless of what damage requests it sends.

We're already gathering damage for wl_surface.damage requests, and these are emitted through the QWaylandSurface::damaged(QRegion) signal, so we should use this information in QWaylandQuickItem to make sure we don't needlessly repaint the entire client if it's more efficient to redraw only parts of it.

Note that some clients/drivers don't submit partial damage if only wl_surface.damage is supported and not wl_surface.damage_buffer (see QTBUG-74927).