Add local SenseVoice recognition service

[LauraGPT]

Summary

• add an offline SenseVoice recognition service backed by sherpa-onnx OfflineRecognizer
• route local microphone recognition to the selected model without silently falling back to the remote API
• centralize complete SenseVoice model discovery (model.onnx + tokens.txt) for startup, updates, settings, and runtime use
• keep SenseVoice on VAD-completed segments and emit final results with capture metadata
• surface local decode failures through the existing service error UI

Validation

• dotnet test tests/VRCLS.Services.Tests/VRCLS.Services.Tests.csproj --nologo --verbosity minimal (13 passed)
• dotnet build VRCLS.sln -p:EnableWindowsTargeting=true --nologo --no-incremental --verbosity minimal (0 errors; existing generated OpenVR nullable warnings only)
• git diff --check
• real sherpa-onnx 1.13.4 decode using the official 229 MB int8 SenseVoice model and test_wavs/zh.wav: 开放时间早上9点至下午5点。

Closes #51