🎉 add f16 simd support for ignore nan

[jvdd]

Is 10x faster than already optimized scalar implementation (using ord_transform - see #44 )

[codspeed-hq]

CodSpeed Performance Report

Merging #46 f16_ignore_nan (324d335) will not alter performances.

Summary

🔥 0 improvements
❌ 0 regressions
✅ 156 untouched benchmarks

🆕 12 new benchmarks
⁉️ 0 dropped benchmarks

Benchmarks breakdown

	Benchmark	main	f16_ignore_nan	Change
🆕	scalar_f16_argminmax_in	N/A	4.2 ms	N/A
🆕	scalar_f16_argmin_in	N/A	2.8 ms	N/A
🆕	scalar_f16_argmax_in	N/A	2.8 ms	N/A
🆕	sse_f16_argminmax_in	N/A	603.2 µs	N/A
🆕	sse_f16_argmin_in	N/A	495.8 µs	N/A
🆕	sse_f16_argmax_in	N/A	495.8 µs	N/A
🆕	avx2_f16_argminmax_in	N/A	294.1 µs	N/A
🆕	avx2_f16_argmin_in	N/A	250.6 µs	N/A
🆕	avx2_f16_argmax_in	N/A	250.7 µs	N/A
🆕	impl_f16_argminmax_in	N/A	294.4 µs	N/A
🆕	impl_f16_argmin_in	N/A	250.8 µs	N/A
🆕	impl_f16_argmax_in	N/A	250.9 µs	N/A