Integration testing instead of artifacts

[tristan-f-r]

Currently, we commit non-input test artifacts directly to the repository with no test-time reproduction (#310 (comment)). Instead, we should generate these artifacts at test-time - if we do care about these files not changing, we can still commit the test artifacts and check that they're equivalent to our currently generated test artifacts (following a similar workflow to that of percy's image testing, or, as a SPRAS example, #320.)