Add SearchQA split materialization helper - #65

Summary

This PR adds a reproducible helper for materializing runnable SearchQA splits from the released ID-only manifest.

The repository ships data/searchqa_id_split, while configs/searchqa/default.yaml points to data/searchqa_split. Without a materialization step, users can hit a missing split directory error or end up creating local data manually. This helper fills that gap by resolving manifest IDs against the Hugging Face lucadiliello/searchqa dataset and writing full train/val/test examples.

Changes

• Add scripts/materialize_searchqa.py.
• Preserve the released manifest order for train/val/test.
• Validate missing manifest IDs and duplicate source IDs before writing output.
• Write runnable items.json files containing id, question, context, and answers.
• Add a generated split_manifest.json with source metadata and counts.
• Document the SearchQA materialization command in data/README.md.
• Add a skillopt[searchqa] optional dependency for the Hugging Face datasets package.
• Add unit tests that avoid network access by testing the materialization core with in-memory rows.

Impact

Users can now materialize the SearchQA split expected by the default config with:

python -m pip install 'skillopt[searchqa]'
python scripts/materialize_searchqa.py

The generated output is written to data/searchqa_split, matching configs/searchqa/default.yaml.

Validation

• /home/thomas/SkillOpt/.venv/bin/python -m pytest -q tests/test_materialize_searchqa.py
• /home/thomas/SkillOpt/.venv/bin/python -m pytest -q
• /home/thomas/SkillOpt/.venv/bin/python -m ruff check scripts/materialize_searchqa.py tests/test_materialize_searchqa.py pyproject.toml
• /home/thomas/SkillOpt/.venv/bin/python -m py_compile scripts/materialize_searchqa.py tests/test_materialize_searchqa.py
• git diff --check