Consider using Blake3 (or something similar) as the hash function for large blobs

[zdave-parity]

Blake3 has built-in merklization, allowing for reasonably concise proofs of subranges of a blob from the blob's hash. Having this capability may prove useful down the line even if we don't make use of it immediately.

Main concerns about using Blake3 are:

• Performance vs Blake2.
• Security margin. Blake3 uses fewer rounds and thus has a smaller security margin than Blake2.