fix: make AI label selection reliable

[jwcesign]

Why

The AI label prompt showed an invalid JSON example and parsed model output without validation. A response using Python-style single quotes therefore raised JSONDecodeError and turned the webhook request into an HTTP 500.

Proposed Changes

• Request a deterministic JSON object from the model and validate its schema.
• Treat malformed, empty, or failed AI responses as no label selection instead of failing the webhook.
• Accept only existing repository label names, skip labels already applied, and prevent AI-created labels.
• Add regression tests to CI and remove the unused workflow variable.

Validation

• python -m unittest discover -s tests (6 tests)
• pylint $(git ls-files "*.py") tests --max-line-length=240 (10.00/10)
• python -m compileall -q app.py bots tests
• docker build --platform linux/amd64 -t gitautomator:codex-test -f Dockerfile .
• Python 3.10 tests inside the built image

Release Note

NONE

[gitautomator]

Thanks to your contribution, the maintainers will review it as soon as they can!