nx_revimg_discover.nx

nx_revimg_discover.nx -- REVERSE-IMAGE CROSS-SITE DISCOVERY engine: the "grow the search results" loop. Given an entity's media fingerprints (the SEED set -- e.g. the Diora Baird images nx_media_gather pulls in) and a crawled corpus of (site, image-fingerprint) records, it answers the operator's three asks: 1. WHICH OTHER SITES host this media? -- perceptual match (dHash Hamming <= thresh), so it catches RECOMPRESSED/rescaled reposts, not just byte-identical copies (sha256 can't). 2. WHAT OTHER MEDIA do those host sites have? -- harvest a host's non-seed images = the entity's other media we didn't have yet. 3. GROW -- fold the harvest back into the seed set and re-discover: new media surfaces new host sites. Fingerprints come from nx_phash (dHash) over decoded images; matching reuses nx_simhash_hamming (popcount64 of XOR) -- ZERO new ranking math. Linear scan here; nx_phash_index (BK-tree) is the scale tier. HONEST: harvesting a host's non-seed images is co-location-based (a fan/official page of the entity is mostly entity media); the grow loop's cross-site re-match is what verifies a harvested image is really the entity (it must reappear on another host). license_tier: ORIGINAL