Server-side metadata layer for "one pack per upload, named from CATALOG
identity" (the chosen model). Pure + unit-tested; wires into the import
path next.

- catalogPackIdentity: derive a stable pack key from an exchange set —
  the longest common prefix of cell stems (US5MD1MC/US5MD2NW → "us5md"),
  a single cell → its full stem, no shared prefix → "" (caller falls back
  to the upload filename). slug() keeps it a valid set name.
- buildSetMeta: assemble a pack's SetMeta from baker.ExtractCellMeta
  overlaid with the CATALOG.031 catalogue — catalogue LFIL supplies human
  chart titles and fills coverage bboxes the header lacks; aggregates the
  scale range, producing agency, union bbox, and cell count.
- writeSetMeta/readSetMeta: <pack>.meta.json sidecar beside the archives.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Each uploaded exchange set now becomes its own pack named from its
CATALOG identity, with extracted metadata exposed to the library.

- extractZipCells: stop dropping CATALOG.031 (its ".031" ext otherwise
  looked like an ENC update file) — parse it and return the *s57.Catalog.
- handleImport: set=auto|"" derives a per-upload pack name via the
  CATALOG identity (longest common cell prefix → "user-<id>"), uniquified
  against existing packs; falls back to the cells' prefix then "upload".
- bakeAndRegister: after baking, ExtractCellMeta + buildSetMeta (catalogue
  titles + coverage overlaid) and write the <pack>.meta.json sidecar.
- /api/packs merges title/agency/scale-range/cellCount/imported; new
  GET /api/pack/<name> returns the full per-cell detail. Delete cleans up
  the sidecar.

Verified end to end against a real NOAA upload (US5MD1MC → "user-us5md1mc",
"Annapolis Harbor", 1:12000, NOAA): /api/packs + /api/pack both correct.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Complete the chart-library feature on the client: each upload is its own
pack with the metadata the backend extracts, and the in-browser OPFS cell
store is gone (everything server-side).

- chart-service.mjs: importZip()/importZipAndWait() (whole-zip multipart →
  /api/import?set=auto) and packDetail() (GET /api/pack/<name>).
- chart-library.mjs: drop the OPFS store + the in-browser zip cell-selection
  flow (_archive/_selected/importSelected/renderArchiveList/_refreshCharts).
  A .zip uploads whole (server names + bakes one pack); a lone .000 uploads +
  auto-bakes. User packs render from /api/packs with title/scale/agency/counts
  and a coverage preview; per-cell detail (title/scale/edition/date) loads
  lazily from /api/pack/<name>.
- chart-library.view.mjs: retire archiveList; refresh the import hints.
- chartplotter.mjs: remove ChartStore; seed _installed from /api/cells (via
  _renderInstalledSets); reroute share-restore to bake a "user-shared" server
  pack instead of caching cells in OPFS. Delete the dead chart-store.mjs.

Backend verified end-to-end earlier; the library UI + share-restore want an
in-browser check (make serve). go test ./... + web tests + node --check green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CATALOG.031 is often absent or omits titles in real uploads, but the
NOAA master index (catalog.json) is a large, separate artifact that must
not become a dependency of the server binary or this feature.

Title resolution is now purely upload-local: CATALOG.031 LFIL when the
exchange set provides it, otherwise no human title — ExtractCellMeta
leaves Title empty (S-57 headers carry only the cell code), and consumers
fall back to the cell Name. All other metadata (scale/edition/date/agency/
coverage) still comes from each cell's own header, so a catalogue-less
upload is fully described regardless. Adds a no-CATALOG test.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

# Conflicts:
#	web/src/plugins/chart-library.mjs