These are things I’ve noticed when using Ghost, but they can be noticed by non-developers when using common features of the web browser. For example, a direct link to an image could be pasted directly into an email or forum post; in those cases, readers will see the direct image URL.
One counter example is Wikipedia, which generally has meaningful image file names. E.g.
Yes, because a hash of random characters implies the lack of meaning, whereas a sequence number or date implies meaning where meaning might not be.
You could use a hash in the folder name (e.g. /oipq5mh6yf/my-image.jpg) so that the file name remains clean. Another option (for caching purposes only, not for uniqueness) is that Ghost could add a hash as a parameter on the end of the URL automatically, e.g. my-image.jpg?oipq5mh6yf.
Another consideration with handling images from older post revisions is whether the images used in those revisions should still be publicly accessible via their existing URL. If I delete or replace an image, and the old image is still publicly accessible via the same URL, that may not be intended by the author. So there may be value in moving images that are no longer found in current post revisions to a private location.