perf: optimize SIMD check and encode implementations

[DaniPopes]

Replace check algorithm with signed overflow trick for large inputs (≥128 bytes). Add AVX2 check path processing 32 bytes per iteration. Double encode throughput (32→64 bytes/iter) via new AVX2 path.

References:

• http://0x80.pl/notesen/2022-01-17-validating-hex-parse.html

Split from #35, credit to @zerosnacks.

[codspeed-hq]

Merging this PR will degrade performance by 30.01%

⚡ 12 improved benchmarks
❌ 4 regressed benchmarks
✅ 20 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	bench1_32b	321.9 ns	460 ns	-30.01%
⚡	bench5_128k	206.2 µs	151.7 µs	+35.89%
⚡	bench3_2k	3.5 µs	2.8 µs	+26.51%
⚡	bench4_16k	26 µs	19.3 µs	+34.71%
❌	bench2_256b	812.5 ns	915 ns	-11.2%
⚡	bench3_2k	7.8 µs	7.1 µs	+10.14%
❌	bench1_32b	429.4 ns	493.1 ns	-12.9%
⚡	bench3_2k	5.6 µs	4.8 µs	+16.34%
⚡	bench5_128k	314.6 µs	260.1 µs	+20.94%
⚡	bench6_1m	2.5 ms	2.1 ms	+21.24%
⚡	bench4_16k	39.7 µs	32.9 µs	+20.46%
⚡	bench6_1m	2.5 ms	2.1 ms	+21.27%
⚡	bench6_1m	1.6 ms	1.2 ms	+36.07%
❌	bench1_32b	1.3 µs	1.5 µs	-10.07%
⚡	bench4_16k	41.6 µs	34.9 µs	+19.16%
⚡	bench5_128k	312 µs	257.5 µs	+21.17%

---

_{Comparing dani/optimize-simd (852c862) with master (023646a)}