Task: Fix dq entity model drift: update to latest format and regenerate

Mirror Fix refdata entity model drift for ores.dq: regenerate every ores.dq domain_entity and junction model (compass codegen entity list) against current codegen templates. Where a model uses an outdated format/convention rather than the templates having simply moved on, update the model to the current convention before regenerating. Commit the resulting clean tree.

Along the way, apply the same lessons learned fixing refdata's drift: check whether any entity's SQL/domain/messaging layer is explicitly hand-maintained (a doc comment flagging missing paste-block coverage) before doing a full regen – prefer --address-scoped regen for anything so flagged; grep synthetic generators for hardcoded/fabricated soft-FK values (coding scheme codes, lookup-table codes) that don't match what a freshly-recreated default database actually seeds.

• Every ores.dq domain_entity and junction model is regenerated via compass codegen entity generate and shows zero diff on a second regen (idempotent).
• Any model found using an outdated format/convention is updated to the current one before regenerating.
• Any entity whose SQL/domain/messaging layer is hand-maintained (flagged in its own model doc, mirroring party's) is excluded from a full regen and only its safe facets (e.g. Qt) are regenerated, same as party's treatment in the refdata task.
• Synthetic generators for FK-referencing fields are checked against actual seed SQL (the default db recreate chain, not optional artefact/dataset scripts) and fixed if they fabricate values that don't exist there.
• compass build and full rat pass after the regenerated files are committed.
• No entity outside dq is touched.

(Implementation strategy. Written when work starts; key decisions are distilled into the parent story's * Decisions at close, but the plan itself stays — it is the historical record of what we did.)

Manual QA scenarios (scaffolded via compass add test_scenario, run through the QA Validation Runner panel) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

All ores.dq domain_entity and junction models regenerated clean. Along the way, uncovered and fixed several codegen infrastructure gaps that were classes of bugs rather than one-off drift:

• Junctions had no <<paste:UUID>> mechanism at all (unlike domain_entity). Added it to the nats-handler and protocol junction templates and to load_org_junction_model=/=_substitute_paste_markers, and used it to restore dataset_bundle_member's hand-written plain list() operation. Its by_bundle operation instead moved onto the existing generic :list_by: mechanism (already proven by calendar_date) rather than a paste block, per "paste blocks only for what isn't already generic".
• A junction or domain_entity with no ** Qt drawer used to abort the entire generate the moment it reached the first Qt unit (junction), or silently emit broken Qt output – empty #include, empty class names (domain_entity, no guard at all) – starving every facet ordered after Qt (repository, service, SQL) of ever being generated. Both are now skipped cleanly under the default full generate; an explicit --address ores.cpp.qt still fails loud. Removed the resulting untracked, never-successfully-generated Qt scaffolding for catalog, subject_area, change_reason, change_reason_category, data_domain.
• Composite text primary keys: the generated NATS handler's history()=/=remove() only ever passed the first pk column to the service, silently dropping the rest of a compound key (caught via subject_area's name=+=domain_name key). Added history_request_extra_args=/=delete_request_extra_args to carry the remaining columns.
• Junction generators had no per-row uniqueness guarantee (unlike domain_entity's counter/idx mechanism) – batch-generating a junction with plain faker-word left/right codes could (and did) collide on the composite PK. Added the same counter/idx pattern to the junction generator template.
• Model-level fixes: stale pre-split ores.dq/domain/... / ores.dq/messaging/... include paths in code_domain=/ =dataset_bundle=/=badge_severity's Qt drawers; an invalid Icon enum value on two entities (Award, Grid – neither exists); change_reason.org's own auto-generated validation function collided by name with the pre-existing, widely-used shared ores_dq_validate_change_reason_fn helper (removed the entity's redundant * Validation function section – nothing referenced it); the dataset_bundle seed upsert function's ON CONFLICT clause didn't match the entity's actual (tenant_id, code, name) composite natural key.
• code_domain's Qt regen was reverted: its BadgeCache passthrough (Badge Mappings tab) has no template support and isn't common to other entities, so it stays hand-maintained until a Qt paste-block seam exists – same treatment as party's SQL in the refdata task.
• Updated stale test expectations (generator version=/ =change_reason_code defaults) and hand-written call sites (registrar.cpp, PublishBundleWizard.cpp, DataLibrarianWindow.cpp, DatasetBundleDetailDialog.cpp) to match the regenerated shapes.

Verification: local build clean (linux-clang-debug-make); full rat (build + ctest across every component) passed, after a db recreate (the schema had drifted: the change_reason_fn name collision and the dataset_bundle seed ON CONFLICT mismatch both blocked schema setup).

Task-ID: B8FE7DA1-18D7-4F33-8402-21411A51A8E2