test: specs for transparency support

[knarewski]

Background
I'm prototyping libvips-based Morandi on a separate branch in hope of improving its memory profile and performance.

Problems
The transparency support is not tested and therefore subject to regressions

Solutions
Add basic test coverage for transparency

Notes

• I added two test cases, because currently morandi handles transparency differently depending on presence of straighten option
• The other differently handled option is cropping outside of image boundaries, where the image is currently blended with white; I decided it's sufficient to stay at 2 test cases though to show that the behaviour can be surprising
• I also expanded visual image matcher to suggest the relevant command when the reference image doesn't exist

Screenshot

A screenshot showing the outputs of new specs (and the inconsistency in handling transparency)

[MrLukeSmith]

It's an interesting behaviour difference ( although possibly expected, I'm not actually sure what "straighten" should do under the hood 😅 ).

[knarewski]

Thanks for review!

I believe it's just an accidental side effect of our implementation of straighten :) . The operation is rotating image by a given angle, then zooming in to fill the entire target area (to avoid leaving black background created in the corners by rotation). We're then "repainting" the relevant part of the zoomed image onto a new image. Perhaps this operation is what applies the transparency instead of stripping it :)