Add feature kyber for Round 3 Kyber API

[jschneider-bensch]

This PR fixes #325, by making available kyber{512,768,1024} modules with their own encapsulate and decapsulate behind the kyber feature, in addition to the regular ML-KEM API.
The differences in Kyber compared to ML-KEM are

• a hash of the initial encapsulation randomness is used instead of using the randomness directly
• the derivation of the shared secret includes a hash of the ciphertext

c.f. FIPS 203 (Draft), section 1.3.

Further, the hybrid KEMs in libcrux-kem are now available in both variants as well.

[franziskuskiefer]

I need to figure out how to make eurydice happy here. But I left some comments to be addressed.

[franziskuskiefer]

With this change C extraction should work again. Looks like eurydice prefers the trait bounds on the functions. Let's do that for now.

[franziskuskiefer]

Thanks, lgtm. But we should also update the C code. And it looks like that extraction is broken again 😬