ASR models (on-device voice search)

Hosts the on-device speech-to-text model Vela downloads for voice search (tier-1, no third-party app or account needed). vela-asr-whisper-tiny.tar.bz2 = Whisper tiny multilingual (int8) + Silero VAD, ~47 MB. Infrastructure release (prerelease so it stays off the latest-release channel); the app fetches the asset on demand. Not an app build.