This adds checks validating that each of the serializers benchmarked (with the exception of some of the zero-copy ones that have less obvious routes back to the original struct type).
we also enable "float_roundtrip" in serde_json to make it pass the test; operating without this feature is a tripping hazard, and i feel like "actually the numbers can just be whatever" isn't a good scope for benchmarking without explicitly setting out to target that use case. frankly i'm kind of appalled that it's not default.
This adds checks validating that each of the serializers benchmarked (with the exception of some of the zero-copy ones that have less obvious routes back to the original struct type).
we also enable "float_roundtrip" in serde_json to make it pass the test; operating without this feature is a tripping hazard, and i feel like "actually the numbers can just be whatever" isn't a good scope for benchmarking without explicitly setting out to target that use case. frankly i'm kind of appalled that it's not default.