Download a model asset, following HTTP redirects (HuggingFace resolve URLs
302 to a CDN), and return its bytes. Never throws.
The bytes are persisted in the MODEL_CACHE_NAME Cache Storage bucket
keyed by url, so a page reload serves the (large) model from disk instead
of re-downloading it. Caching is best-effort: a private-mode/quota failure
still returns the freshly downloaded bytes.
Download a model asset, following HTTP redirects (HuggingFace
resolveURLs 302 to a CDN), and return its bytes. Never throws.The bytes are persisted in the MODEL_CACHE_NAME Cache Storage bucket keyed by
url, so a page reload serves the (large) model from disk instead of re-downloading it. Caching is best-effort: a private-mode/quota failure still returns the freshly downloaded bytes.