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 the application's retry telemetry. When using such a library, run one blank, locally generated input before caching readiness, propagate failures, and reset the shared promise after rejection. Test resolved initialization followed by failed inference, not just rejected initialization. The blank input should require no camera access or personal data.

Applies when: a browser model library treats initialization as best effort. Keywords: browser model preload initialize swallowed errors inference retry