feat(retrieval): page-level BM25 + SPLADE retrieval over OCR'd reports

Self-contained PEP 723 script that indexes OCR'd .mmd reports page by page
and ranks pages for a query, to measure whether a method retrieves the page
containing a given KPI. BM25 + SPLADE on one pyserini/Lucene stack sharing a
single docid space; ColBERTv2 deferred to a later script.

Author: chicham