perf: improve to_base_le

[DaniPopes]

• Closes base_convert: If we keep track of leading zero limbs we can half iterations. #69.
  • This is not a performance gain as the compiler won't vectorize the loop among other things

Based on #488.

[codspeed-hq]

CodSpeed Performance Report

Merging #497 will improve performances by ×2.1

_{Comparing DaniPopes:perf-to-le (130b916) with main (ae691f1)}

Summary

⚡ 6 improvements
✅ 289 untouched benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
⚡	base_convert/4096/4096/10/be	133.4 ms	65.4 ms	×2
⚡	base_convert/4096/4096/10/le	129.6 ms	61.7 ms	×2.1
⚡	fmt/binary/4096	8.8 ms	5.3 ms	+67.46%
⚡	fmt/decimal/4096	7.8 ms	4.2 ms	+83.08%
⚡	fmt/hex/4096	8.4 ms	4.6 ms	+80.78%
⚡	fmt/octal/4096	7.9 ms	4.4 ms	+81.02%