John Baldwin has been working on BSD for decades. Never met him but read his mails since forever. He knows what he's doing. I did wonder why having an existing in kernel blake2 algorithm with fixed size and at his decision to make that a single duplicate function import to get the new one with variable lengths, when you think the other ones calls can be replaced by a macro wrapping the generic blake2(size) as a lambda, but that aside this looks like a solid well reasoned, upstream and Linux friendly bit of work.
Well done.