feat(expriment): support sve2

[hsqStephenZhang]

What type of PR is this?

feat: sve2

TODO

• support get_nonspace_bits on sve2
• support skip_space
• support skip_container
• support skip_string
• support skip_number
• perf
• update document
• clean up code

Current SVE2 approach is pretty native when dealing with skip_xxx, it does not provide a uniform abstraction, it could be optimized.

[hsqStephenZhang]

first round perf result.

Env

AWS Graviton4 AWS Graviton4 CPU @ 2.8GHz. 2 core, 8G mem. support both neon and sve2.

Result

数据集 (Dataset)	测试方法 (Method)	NEON 耗时	NEON-no-reuse 耗时	SVE 耗时	SVE 较 NEON 性能变化	SVE 较 NEON-no-reuse 性能变化
canada	dom::from_slice	3.2093 ms	3.1694 ms	3.1925 ms	🚀 提升 0.52%	📉 下降 0.73%
	dom::from_slice_use_rawnum	2.0917 ms	1.9499 ms	1.9067 ms	🚀 提升 8.84%	🚀 提升 2.22%
	dom::from_slice_unchecked	3.1483 ms	3.1147 ms	3.1395 ms	🚀 提升 0.28%	📉 下降 0.80%
	to_serde_json_value...	7.9253 ms	8.0397 ms	7.7627 ms	🚀 提升 2.05%	🚀 提升 3.45%
	to_simd_json_value...	5.8398 ms	5.8148 ms	5.6302 ms	🚀 提升 3.59%	🚀 提升 3.17%
citm_catalog	dom::from_slice	955.62 µs	1.2900 ms	984.04 µs	📉 下降 2.97%	🚀 提升 23.72%
	dom::from_slice_use_rawnum	954.32 µs	1.3087 ms	1.0123 ms	📉 下降 6.08%	🚀 提升 22.65%
	dom::from_slice_unchecked	913.30 µs	1.2417 ms	944.36 µs	📉 下降 3.40%	🚀 提升 23.95%
	to_serde_json_value...	2.9651 ms	3.1686 ms	2.8944 ms	🚀 提升 2.38%	🚀 提升 8.65%
	to_simd_json_value...	2.4182 ms	2.6329 ms	2.4045 ms	🚀 提升 0.57%	🚀 提升 8.67%
twitter	dom::from_slice	431.21 µs	496.89 µs	399.92 µs	🚀 提升 7.26%	🚀 提升 19.52%
	dom::from_slice_use_rawnum	430.05 µs	499.49 µs	404.47 µs	🚀 提升 5.95%	🚀 提升 19.02%
	dom::from_slice_unchecked	404.89 µs	470.50 µs	373.06 µs	🚀 提升 7.86%	🚀 提升 20.71%
	to_serde_json_value...	1.3289 ms	1.3923 ms	1.3140 ms	🚀 提升 1.12%	🚀 提升 5.62%
	to_simd_json_value...	1.1346 ms	1.1916 ms	1.1292 ms	🚀 提升 0.48%	🚀 提升 5.24%
github_events	dom::from_slice	40.167 µs	46.907 µs	39.226 µs	🚀 提升 2.34%	🚀 提升 16.37%
	dom::from_slice_use_rawnum	40.317 µs	47.112 µs	39.343 µs	🚀 提升 2.42%	🚀 提升 16.49%
	dom::from_slice_unchecked	38.745 µs	45.386 µs	37.850 µs	🚀 提升 2.31%	🚀 提升 16.60%
	to_serde_json_value...	103.10 µs	108.33 µs	103.52 µs	📉 下降 0.41%	🚀 提升 4.44%
	to_simd_json_value...	85.798 µs	91.738 µs	87.726 µs	📉 下降 2.25%	🚀 提升 4.37%
book	dom::from_slice	661.91 ns	619.99 ns	631.25 ns	🚀 提升 4.63%	📉 下降 1.82%
	dom::from_slice_use_rawnum	683.71 ns	629.37 ns	639.21 ns	🚀 提升 6.51%	📉 下降 1.56%
	dom::from_slice_unchecked	640.34 ns	600.24 ns	612.07 ns	🚀 提升 4.41%	📉 下降 1.97%
	to_serde_json_value...	1.7753 µs	1.7690 µs	1.8290 µs	📉 下降 3.02%	📉 下降 3.39%
	to_simd_json_value...	1.2165 µs	1.2083 µs	1.2566 µs	📉 下降 3.30%	📉 下降 4.00%

Explanation

1. There are three round, NEON-no-reuse standards for the cpp version where the skip_space did not reuse the bitmask. the NEON standards for the current approach, and SVE is what this PR does.
2. The current performance is only affected by implementing skip_space on sve2.
3. The performance is very similar to the one in arm: optimize decoder on Arm SVE2 platform bytedance/sonic-cpp#92.