Summary

This adds the complete SenseVoice path requested in the review of block#2520, so the reported Korean locale no longer selects a model that does not support Korean.

What changed

• Adds the official sherpa-onnx SenseVoiceSmall int8 artifact as sensevoice.
• Adds SttFamily::SenseVoice and configures sherpa-onnx 1.13.4 with model.int8.onnx, language = "auto", and ITN enabled.
• Replaces the broad multilingual fallback with explicit primary-language routing:
  • zh, yue, ja, ko -> SenseVoice
  • Parakeet's declared European language tags -> Parakeet v3
  • en and unsupported locales -> the existing English default
• Uses the FunASR Model Open Source License 1.1 notice for SenseVoice instead of the Parakeet CC-BY notice.

Artifact verification

• Archive: sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2024-07-17.tar.bz2
• Size: 163002883 bytes
• SHA-256: 7d1efa2138a65b0b488df37f8b89e3d91a60676e416f515b952358d83dfd347e
• Required files: model.int8.onnx, tokens.txt

I ran the official packaged zh/en/ja/ko/yue samples on CPU with sherpa-onnx 1.13.4. All five decoded successfully; the 4.608 s Korean sample completed in 0.211 s and was detected as <|ko|>.

Validation

• Regression test RED before implementation: ko-KR selected parakeet-v3; ar-SA also selected parakeet-v3.
• cargo test --manifest-path desktop/src-tauri/Cargo.toml huddle:: -- --nocapture: 102 passed, 0 failed.
• cargo clippy --manifest-path desktop/src-tauri/Cargo.toml -- -D warnings: passed.
• cargo fmt --manifest-path desktop/src-tauri/Cargo.toml --check: passed.
• git diff --check: passed.

This PR targets the contributor branch directly so it can be merged into block#2520 without rebasing or cherry-picking.