Story: Classification catalogs as data

The asset class and series subclass classification vocabularies become refdata catalogs that the database owns. The C++ enums that mirrored them are deleted. Every consumer reads the codes and their display names from the catalog tables. Code never duplicates a vocabulary that a table already holds.

• refdata.asset_class_code holds the seven-class taxonomy from the analysis (interest_rates, inflation, fx, equity, commodity, credit, bond); rates is renamed and cross_asset is gone.
• A new refdata.series_subclass_code catalog exists end to end: table, create/drop/populate SQL, domain types, repository, service, protocol, registrar, history mapper, and Qt client.
• The ores.marketdata C++ enums asset_class and series_subclass are deleted; consumers carry the code string and read labels from the catalog.
• Foreign keys from market data, and from trading where it is ready, reference the catalogs.
• Full local build clean and the full ctest suite green.
• Codegen drift stays at zero for both components; entity regeneration reproduces the checked-in output byte for byte.

• Classification is data, not code. The database owns the vocabularies. No C++ enum mirrors a catalog that a table already holds. This rejects the earlier compile-time exhaustiveness approach in favour of existence-based processing: the code carries the code string and reads labels from the table.
• The taxonomy authority is doc/knowledge/domain/asset_classes.org, now a flat seven-class list. rates becomes interest_rates; cross_asset is dropped and its instruments are reclassified.
• series_subclass_code lives in refdata, not in marketdata, because it is a shared vocabulary.

• ores.trading classification changes. A separate workstream owns those, and the trading model is being reshaped at the same time.
• The market data values themselves (quotes, fixings, curves). Only their classification is in scope.