Story: Clean ores.dq to the component clean standard

ores.dq meets the Component Clean Standard: every item: All, Protocol and Entity pass, each item that does not apply is recorded with its reason, and ores.dq joins COMPONENTS_UNDER_TEST. Order 30 of 35 in the sprint 26 component clean-up, tier 3 (domain component).

The component needs two passes. The first gives a model to the SQL that has none, because the standard reads a tree that regenerates cleanly. The second runs the checklist over that tree. The analysis behind the split is the investigation.

• Every checklist item that applies to a component of kind Entity passes, with its evidence recorded on the task.
• Every item that does not apply is recorded as not applicable, with the reason.
• Every generator defect found is fixed at its source, with a regression test.
• ores.dq is listed in COMPONENTS_UNDER_TEST, and check_component_drift.py --all passes.

Pass 1 models the handcrafted SQL. Pass 2 runs the standard. Pass 1 is first, because pass 2 verifies a tree that regenerates.

• Survey on 2026-09-23: 16 entity, junction or operation models; 614 C++ files outside tests, 1 of them generated (0%); 84 messaging headers, 84 of them hand-written.
• Bound to profiles in sprint 25, but not regenerated since the generated marker landed (#2087), so almost no file carries the marker (G03).
• Analysis on 2026-09-26, recorded in the investigation: create/dq declares 74 tables, 49 of them artefact tables; 84 handcrafted SQL files; drift is 517 files.
• The census ran to ground on 2026-09-26 and is now 41 generated, 1 derivable, 0 divergent, 7 orphan — 49 artefact tables, all of them live, so no table is a deletion candidate. Divergent went to zero because the projection is declared in the model rather than reconciled by hand. Progress is on the pass 1 task.
• The design question is answered: an artefact table generates from the same entity model as its base table, and 19 of the 49 already do. The archetypes drop the audit and bitemporal columns and add dataset_id, and codegen forces the dq_ output prefix for an entity any component owns.
• The largest automation gain is the publication functions. 21 files implement them by hand, and their template is wired to no facet.
• 13 of the 19 generated artefact tables have no entity in the 295-entity registry. The drift check cannot see them. asset_classes is one, and it is live.

• Split the work into two passes. Pass 1 models the handcrafted SQL; pass 2 runs the standard. Rationale: the standard's generation items (G01 to G03) cannot pass while 84 handcrafted SQL files have no model, and running the protocol and shell items first would touch code that pass 1 replaces.
• Pair an artefact table with its base by entity name, not by prefix. DQ stages entities other components own, so ores_dq_currencies_artefact_tbl pairs with ores_refdata_currencies_tbl. A prefix rule reports false orphans.
• Do not treat the generated marker as proof of a live model. 13 generated tables have no model. Delete a table only after a consumer census, never on the marker.

• The publish-path decoupling. The DQ publish pattern plan moved publish ownership to the target services. This story keeps publish-from-dq working; it does not revisit where the functions live.