feat: optimize + add f16 scalar implementations

[jvdd]

• add scalar f16 ignore nans
  • test scalar f16 implementation
• add argmin & argmax variants for scalar f16 ignore nans
• add argmin & argmax variants for scalar f16 return nans

[codspeed-hq]

CodSpeed Performance Report

Merging #44 opt_impl (dcc1db9) will not alter performances.

Summary

🔥 25 improvements
❌ 0 regressions
✅ 131 untouched benchmarks

🆕 0 new benchmarks
⁉️ 0 dropped benchmarks

Benchmarks breakdown

	Benchmark	main	opt_impl	Change
🔥	sse_u64_argminmax	1.8 ms	1.6 ms	13.84%
🔥	sse_u64_argmin	1.5 ms	1.3 ms	16.98%
🔥	avx2_u64_argminmax	868.5 µs	825.8 µs	4.92%
🔥	avx2_u64_argmin	740.2 µs	697.5 µs	5.77%
🔥	impl_u64_argminmax	868.7 µs	826 µs	4.91%
🔥	impl_u64_argmin	868.7 µs	826 µs	4.91%
🔥	impl_u64_argmax	868.7 µs	826 µs	4.91%
🔥	sse_i64_argminmax	1.8 ms	1.5 ms	14.51%
🔥	sse_i64_argmin	1.4 ms	1.2 ms	17.99%
🔥	avx2_i64_argminmax	868.4 µs	825.7 µs	4.92%
🔥	avx2_i64_argmin	740.2 µs	697.5 µs	5.77%
🔥	impl_i64_argminmax	868.7 µs	826 µs	4.92%
🔥	impl_i64_argmin	868.7 µs	826 µs	4.92%
🔥	impl_i64_argmax	868.7 µs	826 µs	4.92%
🔥	scalar_f16_argmin_rn	3.3 ms	3 ms	10.24%
🔥	scalar_f16_argmax_rn	3.3 ms	3 ms	10.24%
🔥	sse_f64_argminmax_rn	2.3 ms	2 ms	14.72%
🔥	sse_f64_argmin_rn	2.3 ms	2 ms	14.72%
🔥	sse_f64_argmax_rn	2.3 ms	2 ms	14.72%
🔥	avx2_f64_argminmax_rn	1.1 ms	1 ms	3.95%
🔥	avx2_f64_argmin_rn	1.1 ms	1 ms	3.95%
🔥	avx2_f64_argmax_rn	1.1 ms	1 ms	3.95%
🔥	impl_f64_argminmax_rn	1.1 ms	1 ms	3.95%
🔥	impl_f64_argmin_rn	1.1 ms	1 ms	3.95%
🔥	impl_f64_argmax_rn	1.1 ms	1 ms	3.95%