k12: new implementation with parallel processing support

[newpavlov]

New implementation supports parallel processing of data and consumes data as soon as possible, which allows to remove the 8 KiB buffer used previously. It also introduces separate customized types with owned and borrowed customization strings, while the base type can not be customized.

Potential additional improvements:

• Optimize absorption of chaining values by handling u64s instead of going through the byte buffer.
• Use custom buffering based on consumed_len instead of separate cursors.

Closes #797

[newpavlov]

Benchmark results on my x86 laptop:

$ cargo bench

test kt128_1_10   ... bench:          51.75 ns/iter (+/- 4.34) = 196 MB/s
test kt128_2_100  ... bench:         146.51 ns/iter (+/- 10.80) = 684 MB/s
test kt128_3_1k   ... bench:       1,068.34 ns/iter (+/- 12.90) = 936 MB/s
test kt128_4_10k  ... bench:       9,993.64 ns/iter (+/- 66.05) = 1000 MB/s
test kt128_5_100k ... bench:      99,155.64 ns/iter (+/- 462.52) = 1008 MB/s
test kt128_6_1m   ... bench:     994,643.40 ns/iter (+/- 338,104.49) = 1005 MB/s

test kt256_1_10   ... bench:          55.38 ns/iter (+/- 7.04) = 181 MB/s
test kt256_2_100  ... bench:         171.64 ns/iter (+/- 6.93) = 584 MB/s
test kt256_3_1k   ... bench:       1,310.92 ns/iter (+/- 20.87) = 763 MB/s
test kt256_4_10k  ... bench:      12,423.70 ns/iter (+/- 2,534.85) = 804 MB/s
test kt256_5_100k ... bench:     123,474.30 ns/iter (+/- 880.00) = 809 MB/s
test kt256_6_1m   ... bench:   1,236,551.90 ns/iter (+/- 15,772.23) = 808 MB/s

$ RUSTFLAGS='--cfg keccak_backend="simd256" -C target-cpu=native' cargo bench

test kt128_1_10   ... bench:          53.88 ns/iter (+/- 9.00) = 188 MB/s
test kt128_2_100  ... bench:         137.11 ns/iter (+/- 20.29) = 729 MB/s
test kt128_3_1k   ... bench:         934.46 ns/iter (+/- 8.60) = 1070 MB/s
test kt128_4_10k  ... bench:       8,807.24 ns/iter (+/- 45.38) = 1135 MB/s
test kt128_5_100k ... bench:      61,022.62 ns/iter (+/- 535.21) = 1638 MB/s
test kt128_6_1m   ... bench:     515,211.50 ns/iter (+/- 4,925.42) = 1940 MB/s

test kt256_1_10   ... bench:          51.96 ns/iter (+/- 1.60) = 196 MB/s
test kt256_2_100  ... bench:         158.18 ns/iter (+/- 16.60) = 632 MB/s
test kt256_3_1k   ... bench:       1,155.59 ns/iter (+/- 16.73) = 865 MB/s
test kt256_4_10k  ... bench:      10,914.95 ns/iter (+/- 72.48) = 916 MB/s
test kt256_5_100k ... bench:      76,398.50 ns/iter (+/- 15,408.53) = 1308 MB/s
test kt256_6_1m   ... bench:     631,593.60 ns/iter (+/- 10,271.26) = 1583 MB/s

On M4:

$ RUSTFLAGS='--cfg keccak_backend="soft"' cargo bench

test kt128_1_10   ... bench:          12.64 ns/iter (+/- 0.10) = 833 MB/s
test kt128_2_100  ... bench:          53.46 ns/iter (+/- 0.73) = 1886 MB/s
test kt128_3_1k   ... bench:         458.52 ns/iter (+/- 2.21) = 2183 MB/s
test kt128_4_10k  ... bench:       4,476.62 ns/iter (+/- 36.82) = 2234 MB/s
test kt128_5_100k ... bench:      44,528.55 ns/iter (+/- 200.51) = 2245 MB/s
test kt128_6_1m   ... bench:     446,610.45 ns/iter (+/- 2,878.20) = 2239 MB/s

test kt256_1_10   ... bench:          13.79 ns/iter (+/- 0.08) = 769 MB/s
test kt256_2_100  ... bench:          64.38 ns/iter (+/- 0.31) = 1562 MB/s
test kt256_3_1k   ... bench:         569.71 ns/iter (+/- 6.26) = 1757 MB/s
test kt256_4_10k  ... bench:       5,594.98 ns/iter (+/- 197.50) = 1787 MB/s
test kt256_5_100k ... bench:      55,818.75 ns/iter (+/- 387.30) = 1791 MB/s
test kt256_6_1m   ... bench:     560,366.60 ns/iter (+/- 7,578.58) = 1784 MB/s

$  RUSTFLAGS='--cfg keccak_backend="simd128"' cargo bench

test kt128_1_10   ... bench:          13.82 ns/iter (+/- 0.08) = 769 MB/s
test kt128_2_100  ... bench:          58.81 ns/iter (+/- 0.26) = 1724 MB/s
test kt128_3_1k   ... bench:         506.57 ns/iter (+/- 3.22) = 1976 MB/s
test kt128_4_10k  ... bench:       4,962.32 ns/iter (+/- 26.99) = 2015 MB/s
test kt128_5_100k ... bench:      31,671.00 ns/iter (+/- 677.57) = 3157 MB/s
test kt128_6_1m   ... bench:     293,570.83 ns/iter (+/- 2,292.37) = 3406 MB/s

test kt256_1_10   ... bench:          15.38 ns/iter (+/- 0.14) = 666 MB/s
test kt256_2_100  ... bench:          71.71 ns/iter (+/- 0.75) = 1408 MB/s
test kt256_3_1k   ... bench:         631.36 ns/iter (+/- 6.75) = 1584 MB/s
test kt256_4_10k  ... bench:       6,211.28 ns/iter (+/- 133.55) = 1610 MB/s
test kt256_5_100k ... bench:      39,537.72 ns/iter (+/- 159.29) = 2529 MB/s
test kt256_6_1m   ... bench:     367,704.15 ns/iter (+/- 5,421.26) = 2719 MB/s

$ cargo bench

test kt128_1_10   ... bench:          11.28 ns/iter (+/- 0.10) = 909 MB/s
test kt128_2_100  ... bench:          50.47 ns/iter (+/- 0.74) = 2000 MB/s
test kt128_3_1k   ... bench:         441.50 ns/iter (+/- 5.53) = 2267 MB/s
test kt128_4_10k  ... bench:       4,344.07 ns/iter (+/- 26.11) = 2302 MB/s
test kt128_5_100k ... bench:      24,430.39 ns/iter (+/- 630.76) = 4093 MB/s
test kt128_6_1m   ... bench:     219,602.60 ns/iter (+/- 1,010.68) = 4553 MB/s

test kt256_1_10   ... bench:          12.60 ns/iter (+/- 0.13) = 833 MB/s
test kt256_2_100  ... bench:          62.66 ns/iter (+/- 2.25) = 1612 MB/s
test kt256_3_1k   ... bench:         563.14 ns/iter (+/- 2.60) = 1776 MB/s
test kt256_4_10k  ... bench:       5,588.23 ns/iter (+/- 72.13) = 1789 MB/s
test kt256_5_100k ... bench:      31,290.14 ns/iter (+/- 353.19) = 3195 MB/s
test kt256_6_1m   ... bench:     272,802.07 ns/iter (+/- 2,114.30) = 3665 MB/s