Add fast, byte-oriented Hash derive

[jswrenn]

For types that are IntoBytes, we could derive optimized implementations of Hash.

Adding derive(Hash) to a struct generates a recursive descent into the fields of the struct. Instead, for a type that is IntoBytes, we could relatively easily have derive(zerocopy::Hash) expand to something like this:

impl core::hash::Hash for UserTy
where
    UserTy: ::zerocopy::IntoBytes,
{
    fn hash<H: core::hash::Hasher>(&self, state: &mut H) {
        core::hash::Hasher::write(
            state,
            ::zerocopy::IntoBytes::as_bytes(self)
        )
    }

    fn hash_slice<H: core::hash::Hasher>(data: &[Self], state: &mut H)
    where
        Self: Sized,
    {
        core::hash::Hasher::write(
            state,
            ::zerocopy::IntoBytes::as_bytes(data)
        )
    }
}

Alternatively, if we run with the assumption that Hashers benefit from writing the greatest number of bytes at a time, we can leverage the known size and well-alignedness of the user type to emit calls to state.write_u128, state.write_u64, state.write_u32, and so on. A trailing DST suffix, if any, can be serialized with state.write(trailing.as_bytes()).

We should benchmark these approaches. I don't have a strong intuition for the performance characteristics of Rust's leading Hasher implementations. My guess is that we couldn't do better in the more complicated approach than a well optimized Hasher::write could do, so we're best off doing the simple thing.

[jswrenn]

We should benchmark these approaches. I don't have a strong intuition for the performance characteristics of Rust's leading Hasher implementations. My guess is that we couldn't do better in the more complicated approach than a well optimized Hasher::write could do, so we're best off doing the simple thing.

On further investigation, I don't think there's any benefit to the more complex approach. We should assume that Hasher::write is well-optimized, as it is for:

• ahash: https://github.com/tkaitchuck/aHash/blob/7d5c661a74b12d5bc5448b0b83fdb429190db1a3/src/aes_hash.rs#L152-L208
• fxhash: https://github.com/cbreeden/fxhash/blob/a03cbf42dde010250adef75908f109be197d982b/lib.rs#L99-L120
• libcore: https://github.com/rust-lang/rust/blob/7d40450b2df92bdc9dec414b30cf5f7a5979a92e/library/core/src/hash/sip.rs#L255-L299

[jswrenn]

Partially completed in #2159; still needs a port to 0.9.

[joshlf]

Given that #2159 is labeled needs-backport and given that our 0.9 release roadmap lists needs-backport as a blocker, I think we can close this issue. Reopen if you disagree.