Perhaps someone else can explain it better than me, if so, please leave a comment..
Here's just some general thoughts:
Basically, by decoupling the spriteart from the frame data, we can import/export one image, work on one image without having to flip between frames/animations.
The frame data would remain regardless of whether you add or remove parts of the spritesheet, so you can setup the frame/animation data independently.
It allows management of art assets to be simpler and less exhaustive. You could perhaps have different spritesheets and a sprite can select which to use, or even integrate it with current sprite system and mix them so when you create an animation, you can specify if it uses a spritesheet or not. Then you can simply update/edit the spritesheet and setup the frame data for that animation. You would then be able to edit either of those without affecting the other.
It could allow for modable games, since assets could remain on spritesheets and not be autogenerated into random sheets that mix sprites.
As it is now, the frame data are tied to individual images, so if you remove one it also removes the frame data(imagepoints,collisionpoly), so if you insert another image these are reset. Also you have to flip through them to edit the sprite instead of just remaining on one image and editing it as a whole- so you are required to click and interact with the ui exhaustively. This process becomes a problem if you have many animations with many frames. Games often have many of these, and spritesheets allow say an artist to make many changes to a spritesheet without ever effecting the frame/animation data.
It might also allow for sprites to swap out spritesheets during gameplay. So you can have animation/frame data that can be used with multiple spritesheets, allowing for a character to change its appearance during the game without having to manage extra animation events
This idea seems to mix up the internal representation and the UI representation. They don't have to be the same. For example, the editor could display what looks like a spritesheet, but internally represent frames as individual images; or the other way round, it could display individual images, but internally maintain a single spritesheet image. It doesn't seem clear which you want: the UI improvements, the internal representation, or both? These probably need to be considered separately with a rationale for each approach.