Verify model preload with inference when initialization swallows errors
A model preload is complete only when an inference operation succeeds. Some inference libraries deliberately swallow initialization errors so that a later scan can retry. Caching their resolved initialize promise can therefore report readiness despite missing assets and suppress …