Add evaluation topic benchmark dataset

[tine1117]

Summary

This adds the fixed benchmark topic set requested in #37:

• data/eval_topics.json with 20 topics across CS, medicine, economics, social science, physics, biology, meta-science, and environmental science
• canonical sources for each topic, including DOI or arXiv IDs where available
• expected terms for lightweight coverage checks
• a short schema note in EVALUATION.md

The structure is meant to be easy for the planned evaluation scripts to consume while still being readable for manual review.

Closes #37.

Test

python -m json.tool data\eval_topics.json

Also checked the dataset contains 20 entries and every entry has the required keys.