switch to blake2

Replaced the sha256 hash used internally with blake2b variant which is
much faster in software.  We use the 256 bit version, which is the same
size as sha256.  I was unable to find published test vectors for
blake2b-256, so the b2sum utility was compiled from the blake2 sources,
and the output of that program in 256 bit mode was used to generate
testing data.