nx_asset_meta.nx

nx_asset_meta.nx -- METADATA ENRICHMENT (R3 of the universal organization-tooling arc). Makes the catalog know WHAT each asset IS: it populates the type-specific metadata block plus the Dublin Core core spine ON an asset record, so downstream search/dedup/dashboard can answer "show me 1920x1080 PNGs" or "docs about taxes" without re-opening the bytes (org_research.tsv CONFIRMED: schema-mediaobject = width/height/duration/encodingFormat; dublin-core-core = title/creator/date/ subject; iptc-image-embedded = read intrinsic image facts from the bytes). ZERO NEW STORAGE/IDENTITY -- pure COMPOSITION over what already exists: * record field model + canonical encode/decode + CID -> nx_asset_record (ar_fields/ar_addf/ ar_encode/ar_decode/ar_get/ar_cid/ar_cpy) * REAL image header extraction (width/height/format) -> nx_png_header (png_check_sig / png_parse_ihdr / PngIhdr) [iptc-image-embedded] HONEST EXTRACTION BOUNDARY: * IMAGE (PNG): width / height / encoding_format are REAL-EXTRACTED from the actual bytes via the in-tree PNG header decoder (signature + IHDR). No PNG signature -> not a PNG; we do NOT fabricate dimensions (am_enrich_image returns AM_NOT_PNG and writes nothing -- deny-by-default honesty). * DOC / DATASET: there is no universal binary header carrying a title; the title/subject are PROVIDED (Dublin Core descriptive metadata is authored, not intrinsic). The model + API is the deliverable; real per-format extraction (e.g. JPEG/EXIF, MP4 boxes) wires in exactly here as new am_enrich_* paths, each reusing the same ar_addf record-extension idiom (#19 / #25). Pure/deterministic; no hardware/persistent-firmware writes (Rule 26). license_tier: ORIGINAL