Resolvable vocabulary: build + local content-negotiation deploy

Resolvable vocabulary: build, browsable site, and code-list/data fixes

Builds the second deployment from the architecture doc — the dereferenceable vocabulary served at vocabulary.uncefact.org — and fixes several data/UX issues found while previewing it locally. 7 commits.

Build & local preview

• build-vocab-site.py loads all ontology/**/*.ttl into one graph and emits dist/ — every class, property, code list, code value, collection and domain as index.html + index.ttl + index.jsonld (10,703 resources), plus a landing page, /classes and /code-lists indexes, and whole-graph dumps. Each TTL/JSON-LD is the graph from that point down (a class + its properties, a scheme + its concepts, a concept + its narrower subtree, the root + everything). HTML reuses the existing vocabulary.uncefact.org look & feel, extended for the richer SKOS.
• serve-vocab.py serves dist/ with Accept-header content negotiation (Turtle / JSON-LD / HTML default) — emulating the production CloudFront function.
• methodology/deployment-architecture.md documents the two deployments, negotiation strategy, URL↔file mapping, case-sensitivity handling, and the local → S3 test → prod pipeline.

Code-list navigation UX

• Structured code lists (hierarchy or skos:Collection) now land on their top-level categories with child counts instead of dumping the whole tree; each has a Browse all N codes flat A–Z view. Collections (party-role/collection/*, unit-of-measure/kind/*) are first-class navigable pages. Concept pages give up/down navigation: full ancestor breadcrumb, broader up link, Narrower (down) children, and grouping membership.

Fixes surfaced in preview

• Landing diagram links re-pointed from the docs-site paths (/ontology/<domain>/) to this site — class boxes → their class page, domain regions → /module/<domain>; all links resolve.
• "This resource" links were relative (index.ttl) and resolved to the whole vocabulary; made absolute so they return the sub-graph only. Added page edge margins.
• Case-collision data loss: URIs are case-sensitive but macOS/Windows filesystems are not, so a class (/Product) and same-name property (/product) collided and one silently overwrote the other at build time. Colliding resources now get a distinct on-disk directory recorded in dist/_manifest.json; the server routes through it. S3 keys stay case-sensitive, so production is unaffected.

Vocabulary/data quality

• Renamed 7 terms that collided with a class name by case only (URIs not yet published, so no aliases): address→postalAddress, characteristics→productCharacteristics, measure→measuredValue, party→actingParty, product→suppliedProduct, sensorData→recordedSensorData, and the QUDT dimension annotation → dimensionVector (kept, not removed).
• Deduped language-code-list: the BSP source carried every language twice (uppercase AA + lowercase aa, 185 pairs). Kept the lowercase ISO-639-1 form as canonical and retired the 185 uppercase as deprecated stubs (owl:deprecated + dcterms:isReplacedBy, out of the active scheme but still resolvable). The rule lives in the generator so it's reproducible.

Housekeeping

• Added a repo README (purpose + how to run the browsable ontology locally). Validator (6 gates) and fidelity check pass throughout; the code-list generator reproduces every other list byte-identically.