Add value for BLAKE3 hash function

[tylerwilliams]

Add BLAKE3 hash function.

[tylerwilliams]

@EricBurnett mind taking a look at this? Thank you!

[EricBurnett]

LGTM!

[EdSchouten]

@tylerwilliams I was wondering, what's the reason you folks are pursuing using BLAKE3 and not the recently added SHA256TREE digest function (#235)? Sure, at its peak BLAKE3 is faster than SHA256TREE, but the difference isn't that big in the grand scheme of things, right? SHA256TREE has some added advantages:

• Performance is still adequate on other CPU architectures that provide SHA-256 CPU extensions. For example, BLAKE3 only has 1/10th the throughput of SHA-256 and SHA256TREE on ARM64. BLAKE3 is very strongly opinionated towards x86-64.
• SHA256TREE is built with chunking in mind, allowing you to decompose large CAS objects into smaller ones. BLAKE3 makes this hard, because of their use of Chunk Counters. In Add support for chunking of blobs, using SHA256TREE #233 I'm working towards extending REv2 to provide facilities for that.

I mean, I'm fine with adding support for BLAKE3, but I'm merely interested in seeing whether we can prevent Balkanization in this area.

[tylerwilliams]

@EdSchouten -- Definitely agree re balkanization and I'm very excited to see SHA256TREE land for chunking support. We will likely support this too when it lands in Bazel :)

That said, BLAKE3 is incredibly fast and this is useful when hashing large files, which we do a lot of (docker images, snapshots, large binaries, etc) on CPUs with AVX* support. I don't think it's a bad thing to support this hash in the protocol, or in Bazel, so that these kinds of blobs can be hashed quickly today.