Deterministic ECDSA signing by default?

[jschlyter]

The latest version of cryptography supports deterministic signing for ECDSA. Do we want to provide this in dnspython? By default or as an option?

My proposal is to always sign using deterministic signatures, but could also be persuaded to implement it as an option to dns.dnssec.sign.

[rthalley]

I think we'd want to provide this as it seems generally useful for testing, as you can do test vectors etc instead of testing only "can I verify what I sign". I'm hesitant about having the default be True just for backwards compatibility reasons, though I can't think of why someone would have a test relying on the signatures being different. But experience has shown me that "I can't think of a reason" doesn't mean "it won't break something" :). Also from RFC 6979 I'm left with the impression that the security considerations of abandoning randomness are a little subtle, though I don't see why any of them would matter for DNSSEC. But I'm also not a cryptographer.

[jschlyter]

Draft code in #1104, moving discussion to PR.