Divergences from Photoshop smart-object

Reading https://forums.adobe.com/thread/2372698 to double-check how "smart" a smart-object is, they actually seem quite dumb.

* single image per-smart-object
* resolution needs to match if replacing source graphic (same as with the website that inspired)
* limited actions / transformations applied to smart-object

This does raise questions and potential opportunities.

* this already doesn't care about source resolution. If it's too big it'll get squished, if it's too small, it'll get stretched (that's good)
* because it outputs a layer and is non-destructive, rather than baking into a raster (deferring that for now), layer-effects & filters (to use GIMP terminology) can be applied to the generated image