Task: State what a credit reference entity is

A credit trade states what its reference entity is: the legal entity's LEI, or an identifier of our own data, and the model says which.

• The reference is typed, not guessed. reference_entity_id_type holds LEI or Internal, and reference_entity_id holds the identifier that type describes. A single column holding either could not be joined to anything.
• The ORE import states Internal, because an engine document names its own credit curve, issuer or qualifier and none of those is an LEI. The mapping layer is where a resolver will turn a curve id into an LEI, and the type is what makes that a data problem rather than a guess.
• Resolution replaces the identifier and its type together. A trade whose curve id is resolved states LEI and the LEI; it does not carry both at once.
• The LEI kind has a soft reference to the DQ lei_entity dataset, recorded in the model rather than enforced by a foreign key, as the family's other soft references are.

This task is the column fix only. The two parts it was originally scoped with are their own tasks, because one task is one pull request:

• the synthetic read path into the DQ lei_entity dataset;
• the credit family itself: constructors, a planner, a writer arm, a book and a config, proved by a run.

• A credit instrument states a typed reference, and the type is closed to LEI and Internal.
• A row typed LEI carries a twenty-character identifier, so the string a caller types is refused rather than stored.
• The ORE import states what it actually holds, and its projection tests assert both columns.
• The affected suites are green.

Replace the free-text column with the type and the identifier, regenerate the entity across its SQL, C++ and web facets, and update the one hand-written mapping layer that populates it. The generated fixture states a synthetic LEI so a fixture satisfies the check.

The economic digest covers the economic fields, and renaming one changes every digest that covers it. The tenant that matters holds no credit instruments at all, because the family has never been provisioned: zero rows for ACME. The dev database holds two credit rows, and both are residue from the trading core integration suite rather than customer data; they carry the identifier this change stops generating, and their digests were computed under the old field names. A real environment that holds credit instruments would need a digest migration with this rename, and there is none to migrate here.

The generated shell recipe for credit instruments states __none__ for string columns, including the checked ones. That predates this change – the same placeholder sits in trade_type_code, which has always been checked – so it is recorded as a capture rather than fixed in this task.

A credit trade states what its reference entity is. The column is typed, the LEI case is length-checked, and the ORE import says Internal because that is what an engine document gives us.