Update meshopt_decoder module to latest

[zeux]

This brings up meshopt_decoder modules to the latest versions, which
mostly constitutes a few internal Wasm changes that get reflected in the
different Wasm blobs, and support for Web Workers.

Note that with this change, GLTFLoader will start using decodeGltfBufferAsync
path (see #24460) but by default that still uses the same flow (so no change!).
However, this allows the application to call

MeshoptDecoder.useWorkers(4);

... to activate asynchronous geometry decoding. It's expected that the
better default is to not use WebWorkers, but using them can be
beneficial for very large files or very latency-sensitive applications.

Some size/perf stats:

old decoder (0.14 from before this change):

• .js 21414 bytes
• .js.gz 5745 bytes

new decoder:

• .js 24850 bytes (+16%)
• .js.gz 6442 bytes (+12%, or +700 bytes downloaded)
• 3-4% faster decoding in Chrome
• support for Web Workers (which is where most of the size impact comes from!)

new decoder with worker support stripped (not part of this PR!):

• .js 22715 bytes (+6%)
• .js.gz 5831 bytes (+1%)
• 3-4% faster decoding in Chrome

I'm currently expecting that the extra ~700 bytes of download cost is
Not A Big Deal, and that size-conscious applications use webpack which
may or may not be able to strip Web Worker support out if that is not
used (that said I haven't tried).

[donmccurdy]

I'm currently expecting that the extra ~700 bytes of download cost is
Not A Big Deal, and that size-conscious applications use webpack which
may or may not be able to strip Web Worker support out...

In general we've found it harder than expected to get dead code eliminated by build tools reliably, and I suspect the detection conditions here are not obvious enough for most build tools.

If (in the future) you are looking for a more reliable way to include optional functionality that won't increase bundle size for clients who don't use it, without publishing multiple npm packages, the use of package.json#exports might be the way to go. For example:

import { MeshoptDecoder } from 'meshoptimizer/worker';

But I definitely agree that 700 bytes for web worker support is reasonable, and personally wouldn't bother with additional modules or workarounds in this case.

[mrdoob]

Thanks!