fix: cache gltf model loading

[furoku]

Summary

• add a shared GLTF/GLB model cache so duplicate model URLs reuse one loader request
• normalize model cache keys by removing only cachebust_origin while preserving meaningful query params
• return cloned model instances with cloned materials so hover/translucency state stays isolated per component
• reuse the same cache-key normalization in the existing OBJ/WRL loader path

Test Plan

• bun test tests/model-url-cache-key.test.ts tests/load-cached-gltf-model.test.ts with bunfig.toml temporarily moved aside to avoid this VM's failed sharp preload install
• bunx tsc --noEmit --skipLibCheck
• bun run build
• git diff --check

Note: a full local bun test tests run is blocked in this VM because bun install could not download sharp's native binary through the proxy (407 Proxy Authentication Required). Running without the preload avoids sharp but exposes existing unrelated failures in snapshot/faux-board tests, while the focused cache tests pass.

This PR was prepared with AI assistance from Hermes Agent. I reviewed the generated changes and verification results before submitting.

/claim #93

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
3d-viewer	Ready	Preview, Comment	May 12, 2026 5:45am