Parallelize scrypt with rayon

[smessmer]

Implements #79

If p>1, then the different RoMix blocks are calculated in parallel, using Rayon.

Benchmarks

without rayon feature

test scrypt_17_2_4 ... bench: 270,342,994.10 ns/iter (+/- 37,060,120.60)
test scrypt_17_8_1 ... bench: 294,787,189.50 ns/iter (+/- 21,095,956.53)

with rayon feature

test scrypt_17_2_4 ... bench: 116,287,221.00 ns/iter (+/- 12,892,179.24)
test scrypt_17_8_1 ... bench: 290,716,320.30 ns/iter (+/- 13,595,822.07)