Say which model failed to load and offer Try again; don't start on the backup voice

On the live site a dropped connection left setup at "Loading models 21%" forever. The package reports a failed Whisper download from its worker's start-up step (stage 'worker', fatal), which we never caught, and a failed Kokoro only as 'degraded' while it quietly switches to the MMS voice. Both are now marked failed. The setup screen says which model failed in plain words and shows Try again, which reloads only what failed: a fresh voice engine (back on Kokoro) for Whisper, Kokoro or the avatar, Gemma on its own. Finished files come from OPFS. The interview cannot start on the backup voice.

Also: the microphone notice no longer repeats the package's raw message, and phone-sized touch screens get one line on the home page saying the app is built for a laptop.