feat(spaceship-titanic): add evaluator

[skyzer]

Summary

• add a bundled evaluator for the spaceship-titanic task
• score Kaggle-style PassengerId,Transported CSV submissions against private labels
• write the canonical results.json artifact with accuracy, counts, extras, and per-passenger details
• add regression tests for the evaluator contract and missing/invalid/extra predictions

Why

spaceship-titanic ships with public train/test data, a sample submission, private labels, and a reference agent, but no evaluate.py. That means SIA can run the task but has no first-party evaluator to feed scores back into the generation loop.

Test plan

• python3 -m pytest tests/test_spaceship_titanic_evaluator.py -q
• local CLI smoke test against bundled sample_submission.csv
• python3 -m ruff check .
• python3 -m pytest -q