Story: Fix the has_parent_id feature namespace so profile bindings take effect

A model must state only what it decides for itself. The binding program exists so the bound profile supplies the rest. has_parent_id was the one feature where that did not hold. The profile merge filed the feature under the entity's sql namespace. Every reader looked at the entity root instead. So a profile's assignment never arrived, and each model that needed the feature repeated the line by hand.

The silence hid a second fault. Fifteen entities bound the self-referencing-hierarchy profile, whose contract requires a parent_id_column and a hierarchy_name_field. Only four carried both. The defect alone protected the other eleven. Had the profile assignment ever reached the reader, those eleven would have generated a hierarchy function naming empty columns.

This story moves the feature to the root namespace and repairs what the defect masked. A profile becomes the single source of truth for the hierarchy block, and no model can bind that profile without meeting its contract.

• _FEATURE_NAMESPACE routes has_parent_id to the entity root, so a profile assignment reaches the readers.
• The four hierarchy models (business_unit, counterparty, party, folder) drop their explicit has_parent_id line and still generate their hierarchy function, repository method, service and handler.
• Every model that turns the feature on names both parent_id_column and hierarchy_name_field.
• No model binds self-referencing-hierarchy without those two column properties.
• The eleven flat entities that bound it move to a profile that describes them, and generate no hierarchy code.
• Regeneration over the nine known-drift-free components produces no diff.

• 2026-09-17 (task E2929BA0): fix the namespace, not the readers. Every reader looks at the entity root, so nothing read sql.has_parent_id. Moving the feature to the root is the smaller change, and it matches has_tenant_id and has_workspace_id, which already live there. doc_generate.py corroborated the choice independently: its hierarchical-composite shape preset already grouped has_parent_id with the two tenancy features at the root.
• 2026-09-17 (task E2929BA0): rebind the eleven to uuid-identified-lookup, not uuid-surrogate-lookup. The tiebreaker is the physical-space table. The profile they left and uuid-identified-lookup both enable ores.cpp.eventing-integration-test, and eleven generated integration tests exist for those entities. uuid-surrogate-lookup has no physical-space table at all, so that binding would have deleted them. On feature assignments the two profiles are interchangeable here, because uuid-surrogate-lookup's two extra rows pin flags that already default to false. The fit holds on the descriptive axis too. The reference entity is UUID-keyed, tenant-scoped, parentless and natural-keyed, and ten of the eleven declare a natural key of their own. Only calendar_rule has none, so it alone stretches the profile.
• 2026-09-17 (task E2929BA0): keep parent_id_column and hierarchy_name_field in the root * Flags drawer. They are structural, so no profile can carry them, and the readers already look for them there. Two SQL-facet templates are the only readers, so a later change could move them into * SQL, but that move must change the reader paths too.
• 2026-09-17 (task E2929BA0): has_parent_id belongs to the Keys and columns bundle, not to ores.sql.schema. The templates that read it reach beyond SQL into the C++, protocol and NATS facets. Its narrative home stays on the SQL page, beside the hierarchy block it gates. The feature index, the feature catalogue and the UVL models now file it with the other structural features.

• parent_id_column and hierarchy_name_field stay in the root * Flags drawer while only the SQL facet reads them. The task's Notes record the tidy-up and the change it needs.
• The scaffolder's --entity-<knob> vocabulary. A model authored with compass add entity_org still cannot express the hierarchy block.
• A loader guard that fails when the feature is on without both column properties. It would have caught the eleven at the point of regeneration instead of at the point of audit. Filed as a capture rather than built here.