New SHA2 meet-in-the-middle optimization

[atom]

FWIW we've had this optimization is JtR for a year or so, albeit only in our CPU code yet (not sure why). It was brought up back then by Aleksey Cherepanov in http://www.openwall.com/lists/john-dev/2015/09/23/1.

True, I just wrote it because I didn't find the optimization in the kernels/sha256_kernel.cl source - it wasn't there.