Use unrolled core of blake2 where Argon2 uses the rolled up version

[mheyman]

Argon2.FillBlock.cs:FillBlock() uses the simple version of the inner Blake2 algorithm which can be (in theory) dramatically sped up through unrolling. See Blake2BCore-FullyUnrolled.cs for a version to port.

[mheyman]

In case anybody cares, the "performance" branch has multiple unrolled options and some timing code. So far, I cannot get reasonable timings on my development system (running in a VM) but it does appear that unrolling in managed code doesn't provide a noticeable improvement and may actually slow things down.