Add PyPI publishing workflow

[andimarafioti]

Summary

• Add a tag-driven Publish workflow for PyPI trusted publishing.
• Build and publish the public Linux CUDA 12.8 wheels for x86_64 and aarch64.
• Check wheel metadata with twine check --strict before publishing.
• Add repository release instructions matching the speech-to-speech flow: release PR for version metadata, annotated v* tag, then GitHub Actions publishes through the pypi environment.
• Prepare the first public release as version 0.2.0.

Packaging decision

This publishes only CUDA 12.8 wheels under qwentts-cpp-python for now. The repo still builds CPU and CUDA 13 wheels as validation artifacts, but uploading CPU, CUDA 12.8, and CUDA 13 wheels under the same distribution name/version/platform tag would leave pip with no reliable way to choose the desired runtime.

CUDA 12.8 is the safest default public wheel target because it is the compatibility target we want faster-qwen3-tts users to install by default.

Validation

• Confirmed qwentts-cpp-python is not currently published on PyPI with python3 -m pip index versions qwentts-cpp-python.
• Parsed .github/workflows/publish.yml and .github/workflows/wheels.yml with PyYAML.
• Confirmed publish matrix builds Linux x86_64 and aarch64 CUDA 12.8 wheels.
• Confirmed pyproject.toml version and src/qwentts_cpp/__init__.py __version__ both read 0.2.0.
• uv run --with pytest pytest -q
• git diff --cached --check

Follow-up before first tag

Configure the PyPI trusted publisher for the pypi GitHub environment and the andimarafioti/qwentts-cpp-python repository. Then the first release should be publishable by pushing an annotated v0.2.0 tag from main.