Better testing based on paper: *Type Safety with JSON Subschema*

[Zac-HD]

Type Safety with JSON Subschema is a neat paper which explores similar canonicalization logic to hypothesis-jsonschema, and the implemenation is also in Python!

Unfortunately there are keywords that they don't support in some contexts (see fig. 11), but we should use their tool to test our own canonicalization logic for every schema that both can handle. This is probably more effective than the random-sample-based approach we're currently testing with, though as usual "do both" is likely to be better still. They also have a few corpora of schemas we could probably grab to test against.