Currently if an image is not complete when its rasterized, we will not checker it. This is harder to do than simply skipping the image because the partially loaded image has already been displayed once. So we might need something like using the last decode if its still cached in discardable (looks we can know this at task creation time?), otherwise falling back to the sync decode path. There is still a problem with making sure the update is atomic because the image can be evicted from the cache at any time.