Options for output:

• Rdata containing grid grobs (one per expectation? or test? or context? or package?)
• Raster image: png
• Vector image: pdf/svg

Probably best to store as grid grobs (because has the most information), but if different, display a diff on the binary images (as that's the simplest to calculate and visualise)

Need way to initialise, to say that this graphic looks ok. Also need way to reset - to say that this change is ok, it's not an important difference or it is as expected.

expect_unchanged(name, object)

• name needs to be package unique
• as_expected(name, cache_path)

expect_graphic_unchanged(name, plot)

• if there is a difference, displays in some way - html?