feat: create retrieval benchmark framework with evaluation pipeline and corpus data (#53)

* feat: create retrieval benchmark framework and evaluation pipeline

Add a complete retrieval benchmark package with:
- Evaluation runner and trace system for measuring retrieval quality
- Metrics computation (precision, recall, nDCG, MRR, MAP)
- Strength classification for query-document relevance
- Report generation with formatted results
- Data integrity validation and cross-referencing
- Office document fixtures for testing multi-format extraction
- QREL materialization from retrieval traces
- Source fetching utilities for corpus expansion
- ArXiv corpus expansion command
- MIT lecture import pipeline
- Database benchmark runner for production evaluation
- Controlled corpus ingestion for reproducible testing

* chore: remove stale planning docs and audit artifacts (#54)

Author: thedamod