Adult clinic bundled localization

Asset: site/web/assets/locales/adult-full.js exports ADULT_FULL.
Targets: en zh-CN ja es tl ru fr de vi. Korean is the source language.
All translations were authored offline. No remote translation service is used.

Integration
1. Import ADULT_FULL from ./locales/adult-full.js in web/assets/site-language.js.
2. Merge ADULT_FULL into CATALOG before rebuildPatterns runs.
3. Apply site-language-pattern-order.patch. Specific patterns must run before
   broad equation and unit patterns. The patch also constrains won/hour units.
4. Preserve all input values, echoed user answers, notes and imported original
   document text with data-no-translate. Translate only authored interface text.
5. For menu search compare both Korean keywords and their selected-language
   translation. Search metadata translations are bundled in the catalog.

Rebuild and verify from the project parent directory
python3 adult_clinic_v3/localization/build_catalog.py
ADULT_CHECK_ALL=1 node adult_clinic_v3/localization/verify_catalog.cjs

The verifier reads the shared checkout's actual translation logic. If that file
has not merged the asset yet the verifier binds the catalog in memory and marks
catalog_integration_present false. This verifies the asset without pretending
that the hosted application is already integrated.

For isolating the existing pattern-order issue only
ADULT_CHECK_ALL=1 ADULT_TEST_PATTERN_ORDER=fixed node adult_clinic_v3/localization/verify_catalog.cjs
This modifies only the verifier's in-memory copy. Its report is named
full-verification-proposed-engine-report.json and cannot prove deployment.

Current counts and exact exemptions are in coverage-report.json. It lists every
bundled Korean source key and explicitly retains original academic paper titles
and journal identifiers. source-key-inventory.json maps keys back to data paths.
The input manifest records every TSV checksum for reproducing the build.

Covered learner content: six domains, eighteen skills, twelve baseline items,
seventy-two practice items with feedback, twenty-four followup items, sixteen
intake fields, six decision cases, six learning activities, goals and situations,
resource cards, research findings and their limits, operating guidance and the
curated signup/code/consent source list. Static HTML and curated generated UI
are checked separately in the verification report.

Limits of this asset-only check
No native-speaker review is claimed. Browser screen coverage still belongs to
integration QA. Raw text and CSV exports need an explicit translation path.
PDF/DOCX/HWPX source documents do not become multilingual from this catalog.
Locale-sensitive number entry needs parsing that preserves original input and
canonical values. Original academic author names and titles may stay original.
