Diarization silently fails when only ONNX model is downloaded (expects GGUF)

What happened?

When transcription.recognizeSpeakers is enabled but only the ONNX diarization model has been downloaded, transcription silently falls back to no-speaker-labels output instead of erroring or prompting a re-download. The failure is only visible in the debug log, never in the UI.

Vibe 3.2.2. Model manager state: diar_streaming_sortformer_4spk-v2.1.onnx present in %LOCALAPPDATA%\github.com.thewh1teagle.vibe\; the .q8_0.gguf variant was not.

I expected either:

• The model manager should recognize the ONNX file doesn't satisfy the GGUF requirement and prompt/auto-fetch the correct format, or
• If diarization fails at transcription time, the UI should surface a warning/toast rather than silently producing an unlabeled transcript.

Transcription completes normally with no speaker labels and no visible indication that diarization was skipped. The only evidence is this log line:

gguf_init_from_file: failed to open GGUF file '...github.com.thewh1teagle.vibe/diar_streaming_sortformer_4spk-v2.q8_0.gguf' (No such file or directory) WARN vibe_server::server::diarization: diarization failed, skipping speakers: failed to initialize diarizer ... failed to load GGUF model

I suggest to surface diarization init failures to the UI (toast/banner on the recording), and/or have the model manager verify the GGUF file specifically exists (not just "a diarization model exists in some format") before reporting the feature as ready.

Steps to reproduce

1. Enable transcription.recognizeSpeakers (Settings → Transcription).
2. Have only the ONNX diarization model downloaded (e.g. from an older Vibe version, or a partial/interrupted model download).
3. Record and transcribe a multi-speaker session.
4. Resulting transcript.vibe.json has no per-segment speaker field — diarization did not run.

What OS are you seeing the problem on?

Window

Relevant log output