This allows implementations of the interface to share a single
wlr_buffer for multiple wlr_scene_buffer nodes and still make
the scaled-scene-buffer handle the buffer drops.
wlr_scene_buffer holds an implicit lock while using a wlr_buffer
and scaled-scene-buffer.c holds additional locks for as long as
a buffer is within its internal cache.
This should ensure that a shared wlr_buffer never gets actually
dropped even if wlr_buffer_drop() has been called by another
scaled-scene-buffer instance using the same wlr_buffer.