aead: split `new_test!` into `new_pass_test!` and `new_fail_test!`
This split allows us to skip the pass flags and storing plaintext for failing vectors which gets ignored by run_fail_test.
Unfortunately, this change means that we need to re-generate blb files in downstream crates. (I do not plan to merge this until I generate the new files)
I don’t understand what this PR is supposed to accomplish
Primarily to remove the redundant plaintext field from test vectors which should fail on decryption and to make the passing tests macro a bit simpler. It also would make it easier to see which implementations have failing test vectors and which do not.
Okay, it would be good to have an open PR to AEADs with all of the test vectors converted prior to merging this, so as not to further add to that repo being out-of-sync with this one
If we retain the current test vectors as opposed to the change in this PR, we can add a check that the output buffer does not match the plaintext on decryption failure, which would've caught the bugs in aes-gcm and ascon-aead
I think after we make the final decision on https://github.com/RustCrypto/AEADs/issues/660, we should test it accordingly as part of the test macros, i.e. on decryption failure output buffer must be equal to ciphertext (or fully zeroized). I plan to implement it in a separate PR.
If you do something other than returning the ciphertext, then we would need to change all of the two-pass implementations which abort early to mutate the buffer instead to make the test pass in that case, which I would consider undesirable
@newpavlov I can try to rebase this one if you want.
I guess the best course of action is to split all the blb files according to pass and rework the TestVectors?