Task: Regenerate business_centre from its codegen model and reconcile diffs

Every codegen profile for business_centre produces a zero-diff or explicitly signed-off delta against the repo, per the entity commissioning reference's codegen sync acceptance criteria. In particular: the SQL insert/validate functions carry security definer + set search_path (from the already-fixed sql_schema_create.mustache template) and the bootstrap guard filters active rows; the Qt CRUD (controller, MDI window, detail/history dialogs, client model) lands under ores.qt/refdata (from the model's component=refdata) instead of ores.qt/party, with RefdataPlugin wired to construct/menu it instead of PartyPlugin – no cross-component leakage, achieved via regeneration rather than hand-moving files.

• Output-location check done first for every profile (per the entity commissioning reference): each generated file lands under the correct split-project root (ores.refdata.api / ores.refdata.core / ores.qt/refdata), never the monolith.
• Every generated file is either a zero-diff against the corresponding hand-written file, or has its delta classified and a fix-template / fix-code / accept decision recorded.
• grep for BusinessCentreController in PartyPlugin.cpp returns no construction/menu-wiring hits after reconciliation; RefdataPlugin.cpp constructs the controller and contributes its menu entry.
• Full build (api/core/service/qt.refdata/qt.party/qt.exe) is clean; ores.refdata.core.tests pass; a full compass db recreate applies the regenerated SQL cleanly.
• No generated files for other entities are left modified.

Ran compass codegen entity generate business_centre across every enabled profile and reconciled each diff against the hand-written files it replaced:

• SQL/domain/repository/service/Qt-CRUD split-project output landed correctly (security-definer + bootstrap-guard already came from the template, moved Qt CRUD to ores.qt/refdata with RefdataPlugin wiring instead of PartyPlugin).
• Feature-tested the result (code-run-feature-test: full build, compass db recreate, provision Barclays Plc, a test_scenario doc, both clients driven manually) and used what that surfaced to fix three further gaps: missing has_pagination, coding_scheme_code as a hand-written free-text field instead of a fetch-backed combo (plus a stale FPML_BUSINESS_CENTRE vs seeded FPML_BUSINESS_CENTER spelling mismatch), and no flag icon for country_alpha2_code now that image_id was dropped from the model — the last one needed a new icon_columns accessor (country_flag_icon) since the existing single-flag_icon_column mechanism is image_id-keyed.
• Found and fixed a stale #+image_id: true frontmatter flag that would have resurrected the dropped column on the next regenerate.
• Found and fixed a codegen bug (org_loader.py): insert-trigger Validations rows never carried a column's own :nullable: flag, so a nullable field wired to a validation function had it invoked unconditionally — fixed at the template/loader level, not by hand-patching the generated SQL (business_centre's country_alpha2_code was the first entity to hit this).
• Manual retest of the fixed coding-scheme combo surfaced every code listed twice; traced to iso.coding_schemes=/=fpml.coding_schemes being both seeded unconditionally to the system tenant at DB build time and republished per-tenant via the base=/=crypto dataset bundles during provision tenant — dead weight, since every coding-scheme validator checks only the system tenant. Removed the two bundle-member entries; captured the broader missing system-vs-tenant publish-scope convention as a backlog item (doc/agile/product_backlog/inbox/clarify_system_vs_tenant_dataset_publish_scope.org) rather than fixing it wholesale here.
• Filed the codegen nullable-guard bug as its own backlog capture too (doc/agile/product_backlog/inbox/codegen_nullable_validation_not_guarded.org) since a hand-patch had been applied first before the root cause was traced to org_loader.py, and the fix there is worth a dedicated audit of other nullable+validated fields.

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.

business_centre fully regenerated from its codegen model across every enabled profile (SQL, domain/repository/service, Qt CRUD now under ores.qt/refdata), all diffs reconciled, and feature-tested end to end with the QA Validation Runner scenario passing (Create, Read, Update, Delete, History, Eventing across two clients). Full local build clean (linux-clang-debug-make); ctest 72/72 test binaries passed. Fixed along the way, at their root cause rather than by hand-patching generated output: a stale image_id: true model frontmatter flag, a codegen bug dropping nullable guards on insert-trigger validations, three UX/data gaps surfaced by manual testing (missing pagination, coding-scheme free-text field instead of a validated combo plus a coding-scheme-code spelling mismatch, and a missing country flag icon), and a provisioning bug duplicating coding-scheme rows into every tenant. Two follow-up backlog captures filed for the broader patterns behind the codegen and provisioning bugs.