Add golden tests for the validation benchmarks

[effectfully]

Current the validation benchmarks don't see to have any golden tests ensuring that what scripts evaluate to doesn't change. Unlike e.g. nofib or bitwise benchmarks (and lists ones have some property tests). We should fix that, otherwise an apparent optimization might as well turn out to be a bug.