fix: optimize the ARM function for systems with weak SIMD performance

[lemire]

To optimize on weak ARM cores, I am using a graviton 2 from AWS. It is based on Neoverse N1. The trick is to do the sum across only once per string (or, once per blocks of ~4kB for long strings).

New results on @EgorBo's data

Method	FileName	Mean	Error	StdDev	Speed (GB/s)
SIMDUtf8ValidationRealData	data/Bogatov1069.utf8.txt	510.69 ns	4.810 ns	0.264 ns	2.09
SIMDUtf8ValidationRealData	data/Bogatov136.utf8.txt	92.22 ns	0.438 ns	0.024 ns	1.47
SIMDUtf8ValidationRealData	data/Bogatov286.utf8.txt	176.81 ns	1.814 ns	0.099 ns	1.62
SIMDUtf8ValidationRealData	data/Bogatov527.utf8.txt	284.26 ns	1.160 ns	0.064 ns	1.85

Old results on @EgorBo's data

Method	FileName	Mean	Error	StdDev	Speed (GB/s)
SIMDUtf8ValidationRealData	data/Bogatov1069.utf8.txt	576.68 ns	89.757 ns	4.920 ns	1.85
SIMDUtf8ValidationRealData	data/Bogatov136.utf8.txt	98.05 ns	4.872 ns	0.267 ns	1.39
SIMDUtf8ValidationRealData	data/Bogatov286.utf8.txt	193.19 ns	13.463 ns	0.738 ns	1.48
SIMDUtf8ValidationRealData	data/Bogatov527.utf8.txt	318.96 ns	11.377 ns	0.624 ns	1.65

New results on Twitter:

Method	FileName	Mean	Error	StdDev	Speed (GB/s)
SIMDUtf8ValidationRealData	data/twitter.json	81.39 us	1.671 us	0.092 us	7.76

Old results on Twitter:

Method	FileName	Mean	Error	StdDev	Speed (GB/s)
SIMDUtf8ValidationRealData	data/twitter.json	90.52 us	1.111 us	0.061 us	6.98

New results on Lipsum:

Method	FileName	Mean	Error	StdDev	Speed (GB/s)
SIMDUtf8ValidationRealData	data/Arabic-Lipsum.utf8.txt	33.085 us	0.0820 us	0.0045 us	2.47
SIMDUtf8ValidationRealData	data/Chinese-Lipsum.utf8.txt	28.135 us	0.0356 us	0.0020 us	2.48
SIMDUtf8ValidationRealData	data/Emoji-Lipsum.utf8.txt	26.412 us	0.0459 us	0.0025 us	2.48
SIMDUtf8ValidationRealData	data/Hebrew-Lipsum.utf8.txt	26.829 us	0.2250 us	0.0123 us	2.48
SIMDUtf8ValidationRealData	data/Hindi-Lipsum.utf8.txt	37.910 us	0.1988 us	0.0109 us	2.32
SIMDUtf8ValidationRealData	data/Japanese-Lipsum.utf8.txt	27.758 us	9.4075 us	0.5157 us	2.44
SIMDUtf8ValidationRealData	data/Korean-Lipsum.utf8.txt	26.945 us	1.8570 us	0.1018 us	2.47
SIMDUtf8ValidationRealData	data/Latin-Lipsum.utf8.txt	3.737 us	0.0205 us	0.0011 us	23.26
SIMDUtf8ValidationRealData	data/Russian-Lipsum.utf8.txt	45.083 us	0.0307 us	0.0017 us	2.32

Old results on Lipsum:

Method	FileName	Mean	Error	StdDev	Speed (GB/s)
SIMDUtf8ValidationRealData	data/Arabic-Lipsum.utf8.txt	37.712 us	0.1409 us	0.0077 us	2.17
SIMDUtf8ValidationRealData	data/Chinese-Lipsum.utf8.txt	33.545 us	0.9888 us	0.0542 us	2.08
SIMDUtf8ValidationRealData	data/Emoji-Lipsum.utf8.txt	37.238 us	0.1203 us	0.0066 us	1.76
SIMDUtf8ValidationRealData	data/Hebrew-Lipsum.utf8.txt	33.957 us	0.0790 us	0.0043 us	1.96
SIMDUtf8ValidationRealData	data/Hindi-Lipsum.utf8.txt	44.110 us	0.0708 us	0.0039 us	1.99
SIMDUtf8ValidationRealData	data/Japanese-Lipsum.utf8.txt	32.090 us	1.5966 us	0.0875 us	2.11
SIMDUtf8ValidationRealData	data/Korean-Lipsum.utf8.txt	30.836 us	0.3372 us	0.0185 us	2.16
SIMDUtf8ValidationRealData	data/Latin-Lipsum.utf8.txt	3.687 us	0.1805 us	0.0099 us	23.58
SIMDUtf8ValidationRealData	data/Russian-Lipsum.utf8.txt	50.254 us	3.7907 us	0.2078 us	2.08

[Nick-Nuon]

Looking good to me.

[EgorBo]

I think we can file an API proposal for dotnet/runtime to introduce an API something like:

var hex = v.ToString("X"); // common format symbol for hex

[lemire]

Thanks!!! I did not mean to leave this code there. Removed.

I can file the proposal if you think that's useful that I do it, or I can support it if you do so (if that's useful that I do so).