The current behavior is:
- New images are auto-approved and show up as 'new tests'
- Missing images are just ignored, since no diff is generated for them. But they're still in the 'base image' set
We should instead treat adds/removes as diffs. If needed for data consistency we could even create a white image that's the same size to act as the 'other' in the diff. This would allow us to explicitly approve both removals and additions, and make it more obvious when the test suites change.