Optional Orukeet Core ML speech model for Apple Silicon

[Nathan-Roll1]

I'd like to add Orukeet as an optional local speech model for Apple Silicon users. The implementation is in #980, and I'd appreciate feedback on whether this belongs in FluidVoice's model selector.

The proposed entry is labeled as a Core ML preview. It uses the existing FluidAudio provider, leaves the current defaults alone, and downloads pinned weights from Hugging Face with checksum verification. Installation uses its own cache; transcription stays local. The weights are CC BY-SA 4.0 with Oruk/NVIDIA attribution.

The runtime and installer checks passed on an M5 Max: repeated English, German and French transcriptions, silence, corrupt downloads, cancellation and installation rollback. The PR also fixes incremental download progress. App-specific speed and accuracy ratings remain unset. I'm completing the full app build and GUI evidence separately and will add those results to the PR.

Would you be open to this optional preview, or would you prefer a different integration point?