commit 38028e223d67c7cbc3c8882fe3d6c1408122df98
Author: Haoxiang Fei <feihaoxiang@idea.edu.cn>
Date:   

    perf(normalization): convert the output and scan clean runs faster on JS
    
    - On JS a FixedArray is an array of numbers, so the output buffer's prefix
      converts to a string directly instead of being copied to an exact-length
      array first; FixedArray::unsafe_blit copies element by element there,
      about 3 ns per code unit. The other backends keep the copy inline, since
      out of line it made wasm-gc NFD 10-20% slower.
    - JS skips runs below the clean limit with a sticky regular expression,
      about three times as fast as testing four code units at a time.
    - The wasm-gc block skip compares against U+0200 for NFC instead of ASCII.
    
    Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Changes