feat(audio): add CAMPPlus speaker embedding support - #5298

Summary

• Add the speaker_embedding capability for CAMPPlus audio models, including speaker-vector extraction through /v1/audio/embeddings.
• Add the required ModelScope runtime dependencies so CAMPPlus models can be installed and launched reliably.
• Add a complete Web UI workflow: speech-sample upload, 192-dimensional embedding display, result copying, and “Try To API” examples.
• Fix Web UI ability filtering so speaker_embedding remains selectable.
• Document speaker embedding usage in the Web UI, REST API, and Python client, including cosine-similarity guidance.
• Remove the two CAMPPlus model records from the bundled audio JSON so their metadata can be managed through Xinference Models Hub.

Testing

• Frontend ESLint, TypeScript type check, Prettier, and production build
• Browser QA for the speaker-embedding workflow and responsive layouts
• Pre-commit: Black, Ruff, isort, mypy, and codespell
• JSON validation and Python compilation