feat(importers): add vibevoice-cpp importer for GGUF bundles

[localai-bot]

Summary

• New VibeVoiceCppImporter routes vibevoice.cpp GGUF bundles (e.g. mudler/vibevoice.cpp-models) to the vibevoice-cpp backend, mirroring how the whisper importer handles ggml-*.bin for whisper.cpp.
• Detects by repo name (vibevoice.cpp / vibevoice-cpp), by vibevoice-*.gguf + tokenizer.gguf file pair, or via preferences.backend="vibevoice-cpp". Defaults to the realtime TTS model; preferences.usecase=asr selects the ASR/diarization variant. Bundles the mandatory tokenizer.gguf and (for TTS) a voice prompt, emitting the Options[] (tokenizer=…, voice=…, type=asr) the backend's Load() expects.
• Registered ahead of the existing VibeVoiceImporter so C++ bundles aren't swallowed by the older Python-backend substring match. The Python importer remains the fallback for microsoft/VibeVoice-1.5B and similar safetensors repos.

Example output

For https://huggingface.co/mudler/vibevoice.cpp-models with default preferences:

backend: vibevoice-cpp
parameters:
  model: vibevoice-cpp/<name>/vibevoice-realtime-0.5B-q8_0.gguf
options:
  - tokenizer=vibevoice-cpp/<name>/tokenizer.gguf
  - voice=vibevoice-cpp/<name>/voice-en-Carter_man.gguf
known_usecases: [tts]

preferences.usecase=asr swaps in vibevoice-asr-*.gguf, prepends type=asr, sets transcript, and skips the voice prompt.

/backends/known picks the new backend up automatically since it walks defaultImporters.

Test plan

• go test ./core/gallery/importers/ — 263/263 specs pass (10 new + 253 unchanged)
• Live-HF spec against mudler/vibevoice.cpp-models (TTS + ASR paths)
• Offline fixtures: layout under vibevoice-cpp/<name>/, role auto-detection, quant preference, voice override
• Regression guard: importer does not claim microsoft/VibeVoice-1.5B
• Existing VibeVoiceImporter and TTS-ambiguity tests still pass