It's working as intended. Sorry for the lack of docs. 😬
In the editor, the size of the region attachment's image sets the size of the region attachment. That is remembered for the region attachment. At runtime, if you used a different size image it is stretched over that size. That is a feature, allowing lower or higher resolution images to be used with the same size region attachment. If you change the region attachment image in the editor, it remembers a different size.
All that is without any sequence, but when the region attachment is a sequence, it's the same. The first frame defines the size. In animate mode and at runtime, the images are stretched to that size. You usually want to use the same size for every sequence frame. This allows you to position the frames in relation to each other.
Supporting a different size per sequence frame would take a fair amount of effort. We don't normally load all the images, but we'd need to load them to get their size. We'd also need to monitor them for changes and update the size if they changed. That is a pain to enable different sizes, which isn't really very good anyway: the images would all be centered. If that isn't what you want, you'd have no way to set an origin per frame. It's much better to make all frames the same size, giving you full control over their positioning in a straightforward way. Whitespace stripping removes the extra blank space around the frames when packed into an atlas.
