Use xxhash for byte hashing

[phadej]

Resolves #246

[phadej]

First line is xxhash, second line is master

benchmarking hash/ByteString/strict/5
time                 9.479 ns   (9.433 ns .. 9.546 ns)
time                 18.14 ns   (18.00 ns .. 18.33 ns)

benchmarking hash/ByteString/strict/8
time                 9.590 ns   (9.570 ns .. 9.622 ns)
time                 19.46 ns   (19.36 ns .. 19.58 ns)

benchmarking hash/ByteString/strict/11
time                 9.538 ns   (9.522 ns .. 9.562 ns)
time                 21.19 ns   (21.08 ns .. 21.32 ns)

benchmarking hash/ByteString/strict/40
time                 14.65 ns   (14.62 ns .. 14.69 ns)
time                 50.36 ns   (50.12 ns .. 50.70 ns)

benchmarking hash/ByteString/strict/128
time                 18.68 ns   (18.62 ns .. 18.79 ns)
time                 132.0 ns   (131.9 ns .. 132.2 ns)

benchmarking hash/ByteString/strict/512
time                 40.95 ns   (40.62 ns .. 41.42 ns)
time                 490.4 ns   (490.1 ns .. 490.8 ns)

benchmarking hash/ByteString/strict/2^20
time                 32.28 μs   (32.25 μs .. 32.33 μs)
time                 935.1 μs   (931.4 μs .. 943.7 μs)

the numbers are in the same ballpark as in #246 (comment), absolutely even better. That's maybe my cpu, -march=native and criterion/tasty measurement differences. Anyway, xxhash seems to be clearly better than fnv1, so I'll finish this branch soon.

[phadej]

I'm happy with how this is turning out. Hashing small lazy text and bytestring (<10 bytes) hashing is now slower, but everything else is a lot faster. Hashing small (and large) strict bytestring and text is faster.