Add global coverage audit and multilingual candidate queue

[murraylovecode]

What changed

• audits the verified production and staged corpus across geography, language, role, industry, organization type, source type, and concentration dimensions
• creates a 60-source, locally grounded candidate queue across 7 regions, 33 countries or regional jurisdictions, and 14 language identifiers
• separates accepted, held, and rejected candidates and creates translation, source-identity, role-region, gap-severity, and regional-readiness records
• preserves all non-English translations as staging-only metadata and prohibits translated quotations pending native-language review
• creates no production records or regional pages

Why

The current production corpus is empty and the staged topic corpus is English-only. This batch diagnoses the resulting representation gaps and creates a controlled expansion queue without making claims about regional representativeness.

Validation

• batch schema, geography, language, translation status, decision partition, source identity, duplicates, privacy, and staging isolation checks
• repository data, provenance, attribution, publication, content, duplicate, concentration, and coverage checks
• typecheck: 57 files, zero findings
• tests: 24 passed
• public build: 28 pages and Pagefind index

Review notes

This is a stacked draft PR targeting the BATCH-2026-010 role-comparison branch. Human selection and native-language review remain pending. No candidate is publication-approved.