optimised w+k for digest64

[g11tech]

updated to W64 to W64+K for additional optimization as pointed by @potuz, to save 64 adds and 64 mem loads

Also mystery of variable performance on various machines was resolved:

1. The previous test/bench.js, was generating the random buffer inside the bench loop, causing interference of that as it also involved Math.Random, multiplications and rounds off.
2. The cpu clock turbo boosts should be switched off for true comparison (minor factor in enviorments where there is a fixed cpu clock but causing huge variation where there are turbo boosts as the computation profile of normal digest and digest64 become very different)

pre W+K optimization digest64 performance with bench fixed : ~ 24% faster than normal digest

post W+K optimization: ~ 29% faster than normal digest

GAIN: ~ 5%
@dapplion @wemeetagain
ps: this branch doesn't include uint8 optimization gains to be added by @tuyennhv PR!