feat: add constant-time trait bounds

[AaronFeickert]

Currently, the only implementation of the SecretKey and PublicKey traits is for Ristretto, where both scalars and group elements use constant-time equality in their underlying PartialEq implementations, and which support the ConstantTimeEq trait.

This PR does what it can to encourage the use of constant-time equality for keys by doing a few things.

First, it requires that any types implementing SecretKey or PublicKey also implement ConstantTimeEq. Unfortunately, this doesn't guarantee that their PartialEq implementation defaults to this, and it doesn't appear possible to enforce this at the trait level.

It also sets a good example by manually implementing PartialEq on the Ristretto key types to use their ConstantTimeEq implementations. This isn't strictly necessary, but hopefully helps to indicate best practice. It also implements ConstantTimeEq directly as required by the new trait bounds.

Finally, it implements ConstantTimeEq for DiffieHellmanSharedSecret using the new trait bound, and removes a redundant Zeroize trait bound.

Note that this doesn't actually change the current implementations' behavior, and therefore incurs no performance hit.

Closes #139.

[sdbondi]

utACK

[AaronFeickert]

I have an open PR in subtle to add a marker trait intended to signal that all equality testing for a type is done in constant time. If this is merged and then used in the curve25519-dalek types, it could be used as a trait bound here.