Add corpus.jsonl dataset via git-lfs (478 Wikipedia science docs, 2.5M tokens)

Add .gitignore, dataset metadata, retrieval layer, and latest web/graphrag updates