x/crypto: add new package siphash

[aead]

This is a proposal for implementing the pseudo-random function SipHash.
SipHash was designed to defend hash-flooding DoS attacks, but it can be used in many use cases.

I've seen some projects using SipHash to solve different problems and use custom implementations. So it's probably a good idea to provide a "official" golang implementation.

[gopherbot]

CL https://golang.org/cl/38362 mentions this issue.

[odeke-em]

/cc @agl and other crypto folks.

[agl]

I believe that siphash meets the bar for x/crypto.

[dgryski]

Ping. Seems like @aead was fine to abandon the CL, but @agl seemed in favour. I'm in favour of closing.

[aead]

I think there is no obvious use case for SipHash (anymore). Esp. since a reasonable fast implementation probably needs assembler code (See #28126). Further there is a "successor" (https://github.com/google/highwayhash). We can consider adding SipHash again if it is requested by (other) users?!

[FiloSottile]

I agree with @aead for a few reasons:

• I don't like the idea of having a fast MAC which is not a secure hash but is called Hash in x/crypto
• x/crypto already has too many optimized implementations which we don't have a solid maintenance story for
• there are two main implementations in the ecosystem, so it's not too fragmented https://godoc.org/?q=siphash
• HighwayHash exists now and since SipHash uses are mostly internal, and not protocol-exposed, I'm not sure why one would use SipHash