perf: reimplement optimized natural sorting algorithm, speed up ~6 times for case-insensitive sorting

[sxyazi]

Motivation

This PR fixes the performance regression caused by the introduction of case-insensitive sorting.

Quote from Discord:

I have identified the cause of the performance regression.

It's because the Yazi Git version introduced case-insensitive sorting as the default behavior, and when users set the sorting method to "natural", it uses the natord crate to do it.

The case-insensitive implementation in this crate is very inefficient, and the library hasn't been maintained for 8 years. This has resulted in a doubling of sorting speed in Yazi.

I'm currently working on rewriting a new, optimized sorting algorithm. I'm also considering reverting the default behavior to case-sensitive, since it does have an impact on performance.

Users will only need to manually enable it when they truly require it, so they won't pay the performance cost when it's unnecessary.

BTW, I also stumbled upon this, seems Yazi and exa share the same performance issue 🤣 ogham/exa#367

Benchmark

Sorting 100,000 files with case-insensitive sorting, speeds up by 6 times:

Since Yazi didn't introduce any benchmark framework, I used Instant for a rough measurement, which is simple enough. There is benchmark code in natsort.rs as well; you can run it yourself.