Fixes #8524. Draco glb loading failed in Node: initializeWorkers in draco-decoder.js unconditionally used browser-only APIs (Blob, URL.createObjectURL, the web Worker constructor), and the worker body in draco-worker.js talked to self — none of which exist in a Node worker_threads worker.

This branches on platform.environment === 'node' to use worker_threads, mirroring the existing gsplat-sorter / gsplat-sort-worker pattern already in the engine.

Changes:

• draco-worker.js: resolve the worker scope as const myself = (typeof self !== 'undefined' && self) || require('node:worker_threads').parentPort; and use myself.postMessage / myself.addEventListener('message', …) instead of self. The decoder module factory is resolved as myself.DracoDecoderModule || (typeof module !== 'undefined' && module.exports) — the emscripten glue exposes it as a global in a browser worker but on module.exports (CommonJS) in a Node eval worker.
• draco-decoder.js: in Node, spawn new Worker(code, { eval: true }) and listen with worker.on('message', …); the browser keeps the blob-URL path. The message handler normalizes message.data ?? message (web MessageEvent vs Node raw payload).

Notes:

• As with the existing gsplat sorter, the Node main thread relies on a global Worker being available — consumers shim globalThis.Worker = require('node:worker_threads').Worker.
• Out of scope: downloadScript / compileModule still require http(s)-reachable URLs in Node (local file paths won't fetch).

Verified by decoding examples/assets/models/heart_draco.glb with the modified source in both environments (identical output — indices 1080, vertices 12960, stride 24, position + normal): in Node (environment = node) and in the browser (environment = browser).