primitives: Implement sr25519

[Yawning]

Because it is a better algorithm, and people should use it, implement sr25519. This is a notable improvement over the existing Go implementation that appears to be popular because:

• It is a lot faster in part due to:
  • The underlying operations using a lot more assembly.
  • Not having to re-derive the public key each time the user signs.
  • Instead of calculating the verification by hand (lolololololol), we do it in one-shot.
• It is more correct:
  • The s11n routines are more idiomatic (and probably actually work as intended). (Fixed)
  • This actually samples 128-bit random scalars when doing batch verification. (Fixed)
  • This uses a transcript RNG, instead of YOLO ignoring the private key nonce.
• Support for SigningContext instead of having to work with a merlin transcript.
• A more idiomatic/nicer API:
  • Not having foot+gun API calls like MiniSecretKey.Public().
  • Can override the entropy source wherever one is called for.
  • BinaryMarshaler/BinaryUnmarshaler instead of the Encode/Compress/Decode/whatever.

TODO:

• Write tests for the key generation/s11n.
• Implement batch verification.
• Add license information.

TODO (deferred to a later date):

• (perf) Replace the strobe implementation with something that doesn't guzzle heap.
• (feature) Implement the VRF construct (The w3f repo has an issue hinting that this will change).
• (feature) Implement the HKD construct.

Part of #4