Optimize utf8 decoder

[ycw]

This PR attempts to boost perf for loaders which need decoding utf8 from an uint8array;

The implementation aligns to new TextDecoder('utf-8', { fatal: false }) meaning that malformed data will be substituted with replacement character.

Benchmark: https://jsperf.app/genija/2
in Edge 124.0.0.0 / Windows, it outperforms the old approach #12933 and TextDecoder both by ~90%

Tests: https://jsfiddle.net/sa37f1qh/

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

Filesize dev	Filesize PR	Diff
674.4 kB (167.1 kB)	675.3 kB (167.3 kB)	+841 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

Filesize dev	Filesize PR	Diff
453.8 kB (109.6 kB)	453.8 kB (109.6 kB)	+0 B

[Mugen87]

TBH, I'm not a fan of such implementations when there is a standard API that can be used. I understand there is a performance boost but the new code is also hard to maintain and read. And there are currently no complains about too slow parsing times in context of decodeText().

I would prefer to actually remove the entire TextDecoder fallback in decodeText() and always rely on this class since it should be supported in all common browsers and node. If a project reports compatibility issues, it has to add add a TextDecoder polyfill.

[ycw]

I would prefer to actually remove the entire TextDecoder fallback in decodeText()

done. #28278

---

closed. Reason: hard to maintain