fix(detector): parse labeled AI score

[mikemikimike]

Description

Fixes the AI/LLM probability score parser so unrelated numbers in generated explanations cannot override the model's score.

• Parse scores only from labeled AI_SCORE, Score, or Probability lines.
• Keep the neutral default score of 5 when a valid labeled score is missing or out of range.
• Require Gemini responses to end with a dedicated AI_SCORE line.
• Add regression coverage for score fractions, unrelated numbers, and invalid scores.

Issue: #22

Fixes #22

---

Type of Change

• Bug fix (non-breaking change fixing an issue)
• Tests (adding missing tests or fixing existing tests)

---

Validation

• pytest tests/ (6 passed)
• ruff check .
• python -m py_compile setup.py insight/*.py
• python -m build --sdist --wheel

Checklist

• My code follows the style guidelines of this project (PEP 8 / ruff).
• I have performed a self-review of my own code.
• I have added tests that prove my fix is effective.
• All new and existing tests pass locally.