Add voice reference extraction API

[andimarafioti]

Summary

• Bind qwentts.cpp qt_extract_voice_ref / qt_voice_ref_free through ctypes.
• Add a public VoiceRef API plus .spk / .rvq save/load helpers.
• Add QwenTTS.extract_voice_ref(...) and QwenTTS.save_voice_ref(...) for in-process Base clone latent extraction.
• Bump the package version to 0.3.0 and update HF wheel install examples.

Why

faster-qwen3-tts needs to cache cloned voice conditioning without shelling out to qwentts.cpp tools. This exposes the native ABI directly in the Python package so future wheels include the required API.

Validation

• uv run --with pytest pytest -q tests
• uv run --with ruff ruff check src tests
• uv run python - <<'PY' ... confirmed qwentts_cpp.__version__ == "0.3.0"
• local native loader smoke confirmed qt_extract_voice_ref and qt_voice_ref_free symbols are available