Thank you for the comment! And sorry for the incredibly late reply.
The renpy.image_size function was intentionally ignored for efficiency reasons. The renpy builtin loads and unpacks the full file without making use of any caching. The __get_image_size function only reads the first couple dozen bytes of the file to fetch the width and height from the metadata.
For the use case, I felt that it would be unnecessarily costly to load, decompress, then throw away an image of unknown size just to get at a couple of ints.
If renpy adds a similar method that does make use of image caching, that would the preferable option.