Create fuzz testing for defer/stream

Looking at the schema it should be possible to determine all the positions where `@stream` is valid (and `@defer` is valid across arbitrary fragments generally), and we have snapshots of what the final data should be - so we should be able to run some "fuzzing" where we randomly manipulate our queries to add defer/stream and then play back the incremental delivery stream to build the final payload and compare that to the expected data - it should _always_ match unless errors are expected.