Bump signature to v3.0.0-rc.4

[tarcieri]

Also bumps ecdsa to v0.17.0-rc.7

This brings *DigestSigner/*DigestVerifier changes introduced in RustCrypto/traits#2004

cc @daxpedda

[daxpedda]

@tarcieri should I do a follow-up where I change it to Default + FixedOutput + HashMarker + Update?

[tarcieri]

@daxpedda sure, if you can do that across the board both here and in the signatures repo that'd be great

[daxpedda]

@tarcieri so I did attempt to do this for signatures, but its not that straightforward.
You can find my attempt here: daxpedda/signatures@f89e410.

Basically dsa and ecdsa require rfc6979, which in turn uses SimpleHmacReset, which requires Digest.
So the first thing I did is switch rfc6979 to HmacReset, which "just" requires EagerHash (which still requires Digest).
After that, I started replacing all mentions of Digest in dsa and ecdsa with EagerHash.
This worked for dsa, but ecdsa doesn't use rfc6979 for verifying, which is behind a separate crate feature, so it doesn't pull in hmac either, not giving me access to EagerHash.
Because ecdsa::hazmat::DigestAlgorithm::Digest now requires EagerHash, I would need to pull in hmac for feature = "verifying".

So its all solvable, but the change is pervasive, let me know if you want me to continue with EagerHash.

Alternatively, we could just make Digest a supertrait over Update, among other traits. It always bothered me in the first place that this isn't the case.

I assume I will hit the same snag in elliptic-curves because its built on ecdsa.

[tarcieri]

Perhaps ecdsa has too many knobs. I've been thinking about getting rid of the features which avoid providing an algorithm implementation.

For now though, consolidating into one feature for both signing and verification seems fine to me.

[daxpedda]

What about changing everything from Digest to EagerHash? That's a pretty big change, is that alright?

[tarcieri]

Definitely makes sense for *Signer due to the use of HMAC. I guess I still don't completely understand why it's needed for *Verifier. Is it just for parity with *Signer?

[daxpedda]

Its because both signer and verifier implementations use type ecdsa::hazmat::DigestAlgorithm::Digest. But to make this type work for signer implementations I had to write type Digest: EagerHash and I can't write that if I'm unable to import the type.

[tarcieri]

Aha, makes sense. Sounds fine to change them all to EagerHash then.