Slow with Arabic (esp vs fzy)

- [x] I have the latest version of fzf
- [x] I have read through the manual page (`man fzf`)
- [x] I have searched through the existing issues

## Info

- OS
- [x] Linux
- [x] Mac OS X
- Shell
- [x] bash

## Problem / Steps to reproduce

I appreciate that this is not a common use case however I have noticed that **fzf** is many times slower than **fzy** when searching Arabic (and fzy's results are quite good).

### Dataset
```bash
wget https://dumps.wikimedia.org/arwiki/latest/arwiki-latest-abstract1.xml.gz
gunzip arwiki-latest-abstract1.xml.gz
```

`fzf`
```
$> time cat arwiki-latest-abstract1.xml | fzf -f 'هذه تجربة' | wc -l
6427
real 0m3.425s
user 0m16.836s
sys 0m0.624s
```

`fzy`
```
$> time cat arwiki-latest-abstract1.xml | fzy -e 'هذه تجربة' | wc -l
6479
real 0m0.389s
user 0m1.352s
sys 0m0.267s
```

`+x +i --algo=v1 --literal` really helps, but there remains a big difference
```
$> time cat arwiki-latest-abstract1.xml | fzf +x +i --algo=v1 --literal -f 'هذه تجربة' | wc -l
6427

real 0m1.925s
user 0m2.761s
sys 0m0.494s
```

Start by reproducing the provided Linux and macOS shell benchmarks with the Arabic dataset and compare fzf's timings against fzy. Investigate the fzf filtering entry point and profiling results; done means reducing the Arabic-search performance gap without regressing the documented search behavior.

Written by the indexing model from the issue text.