fix(evaluation): Support direct run_evaluation script execution

[Gujiassh]

Description

The evaluation README tells users to run python run_evaluation.py from the evaluation directory, but run_evaluation.py only used a package-relative import. Direct script execution therefore failed before argument parsing with ImportError: attempted relative import with no known parent package.

This PR keeps the package import path for python -m evaluation.run_evaluation and adds a direct-script import path for the documented python run_evaluation.py invocation.

More Information

• Adds a small __package__ check at the top of evaluation/run_evaluation.py.
• When executed directly, the repository root is added to sys.path and evaluation.utils is imported by absolute package name.
• Package/module execution continues to use the existing relative import.
• Adds a regression test that loads the script with runpy.run_path and stubs optional heavy evaluation dependencies.

Validation

uv sync --extra test --extra evaluation
uv run pytest tests/test_evaluation_entrypoint.py
uv run pytest tests/test_evaluation_entrypoint.py tests/test_cli.py
uv run ruff check evaluation/run_evaluation.py tests/test_evaluation_entrypoint.py
uv run python -m py_compile evaluation/run_evaluation.py tests/test_evaluation_entrypoint.py
git diff --check

Linked Issues

Fixes #371