Faster hex encoding

[novacrazy]

Optimizes the standard encoding, plus adds support for using the faster-hex crate for SIMD-accelerated encoding.

Expand for benchmarks table

N	Before	After	Change
1	16.220 ns	16.010 ns	-1.30%
2	17.035 ns	16.019 ns	-5.96%
4	17.765 ns	17.000 ns	-4.31%
8	18.798 ns	18.640 ns	-0.84%
12	21.002 ns	20.106 ns	-4.27%
15	22.648 ns	21.692 ns	-4.22%
16	22.495 ns	15.583 ns	-30.73%
32	32.556 ns	16.930 ns	-48.00%
64	50.232 ns	19.022 ns	-62.13%
100	70.175 ns	23.459 ns	-66.57%
128	87.228 ns	20.423 ns	-76.59%
160	105.057 ns	22.228 ns	-78.84%
255	156.836 ns	34.361 ns	-78.09%
256	182.728 ns	25.557 ns	-86.01%
500	293.572 ns	41.154 ns	-85.98%
512	352.274 ns	38.409 ns	-89.10%
900	610.794 ns	63.040 ns	-89.68%
1023	585.836 ns	74.845 ns	-87.22%
1024	586.594 ns	65.048 ns	-88.91%
1025	505.389 ns	72.909 ns	-85.57%
2048	975.273 ns	110.846 ns	-88.63%
4096	1945.770 ns	201.676 ns	-89.64%
5000	2343.561 ns	250.155 ns	-89.33%
10000	4698.586 ns	509.036 ns	-89.17%

It's worth noting at least 10-16ns of those benchmarks seems to be spent on non-encoding things. In fact it was made worse when I upgraded to Rust 1.72.0 today, but only in terms of that base offset.

I'm not super happy about the unsafe usage here, but when mixing the fmt precision in like that LLVM has a harder time reasoning with it.

[novacrazy]

Re-ran benchmarks with Rust 1.73.0-beta, which includes an LLVM upgrade:

Expand for benchmarks table

N	Before	After	Change
1	9.153 ns	9.202 ns	0.53%
2	9.820 ns	9.293 ns	-5.37%
4	11.704 ns	9.952 ns	-14.97%
8	13.164 ns	11.583 ns	-12.01%
12	13.737 ns	13.645 ns	-0.67%
15	15.083 ns	14.719 ns	-2.42%
16	16.915 ns	12.643 ns	-25.26%
32	28.896 ns	14.379 ns	-50.24%
64	47.562 ns	15.429 ns	-67.56%
100	75.660 ns	19.488 ns	-74.24%
128	83.917 ns	18.227 ns	-78.28%
160	109.399 ns	23.072 ns	-78.91%
255	156.394 ns	32.404 ns	-79.28%
256	159.777 ns	24.415 ns	-84.72%
500	288.924 ns	39.826 ns	-86.22%
512	296.679 ns	36.177 ns	-87.81%
900	548.176 ns	58.511 ns	-89.33%
1023	583.620 ns	73.875 ns	-87.34%
1024	614.904 ns	63.397 ns	-89.69%
1025	566.742 ns	68.560 ns	-87.90%
2048	1084.768 ns	113.175 ns	-89.57%
4096	2146.710 ns	198.601 ns	-90.75%
5000	3290.758 ns	245.378 ns	-92.54%
10000	5580.791 ns	488.860 ns	-91.24%

Old version got slower, new version gets faster. Neat.