publish_from_dq codegen facet: hand-writing artefact tables is a recurring cost

Implement the publish_from_dq codegen facet (designed but never built — see References) so a DQ-published dataset's artefact table, publish-from-dq SQL function, NATS handler, and registrar wiring are generated from the owning entity's existing codegen model, instead of hand-written per dataset.

Discovered while raising a task to publish a curated FX driver-rate dataset via DQ/librarian, the same way refdata.currency_pairs is published: there is no codegen model anywhere in the tree for a DQ artefact table (confirmed by grep across every modeling/ directory) — dq_currency_pairs_artefact_tbl, the synthetic FX spot configs artefact, and now this new dataset's artefact table all had to be hand-written from scratch, each shape near-identical to the last (dataset_id/tenant_id/version + domain columns, a publish-from-dq SECURITY DEFINER function reading the artefact and upserting into the real domain table). The facet's own design task called for proving it against report_definitions with zero behavioural diff — that reference target and design still stand; there are now three real hand-written instances to generalise from instead of one.

• Design the publish_from_dq codegen facet — ABANDONED.
• Implement the publish_from_dq codegen facet — ABANDONED.
• Publish a curated FX driver-rate dataset for 2016-02-05 — the task that hit this gap a third time.

• Synthetic data librarian support: FX foundation — the story both abandoned facet tasks belonged to.