I am not sure what exactly granularity means in torchstat. As far as I can tell, nothing really changes and when I was trying to look at the class implementations of StatTree and StatNode, I didn't really understand what the purpose for granularity property was. Can someone explain this?