When a guest app crashes or calls
vkDestroyInstance, !501 (closed) sets those remaining syncs belonging to the destroyed context to signal in the timeline, so that the syncs will later be retired known by crosvm.
When a guest app holds the instance but destroys the device and create new ones afterwards, both the
pending_syncs tied to the prior device need to be retired properly.
ASYNC_FENCE_CBis set, we can retire those directly at
ASYNC_FENCE_CBis not set, we destroy the
vkr_queue_destroyand move the queue sync to the context to be retired and freed in the next