nx_search_cli.nx

nx_search_cli.nx -- ASSIGNMENT E (tutored): the search engine's FRONT DOOR over the DURABLE index. The harvest (nx_library_harvest_v2) proved build->persist; until now nothing user-facing QUERIED that artifact -- the index was provably correct and practically dark. This is the live path: argv terms -> load knowledge/index/library.idx (NO rebuild: build-once-query-forever) -> inverted-index postings shortlist the candidate docids -> docmap names their files -> the Engineer-wired bm_* BM25 ranker (x1e6 fixed-point, the _hdl_build nx_bm25.nx that shadows runtime's on this lane) scores them -> "score path" per hit. Distinct from nx_library_search.nx (the Researcher's IN-MEMORY corpus ranker; no main, no persistence) -- one capability per module. BM25 stats (avgdl/IDF) are computed over the CANDIDATE set; corpus-wide stats stored at harvest = the next rung when the corpus outgrows reading candidates whole. Usage: nx_search_cli.sov.elf <term> [term ...] license_tier: ORIGINAL