That's a good point; as I recall reading it it was supposed to say HMAC-SHA256. But, to be clear: HMAC-MD5 isn't really busted, I just would prefer not to see it in new code.

Exactly. We want to move away from md5 (and SHA-1) and has been trying to do that for many years now. So we should not use it _even if_ it is ok in a specific construction. It adds yet another dependency that makes it harder to remove md5.