In case anybody wonders why it changed in 1.10 (like me).
Addressing images by their content also lets us more easily detect if something has already been downloaded. Because we have separated images and layers, you don't have to pull the configurations for every image that was part of the original build chain. We also don't need to create layers for the build instructions that didn't modify the filesystem.
So no change in the FS is no physical layer, only a logical layer and we only need to mind the physical layers.