A closed-set member's external name, apart from its identifier (spike first) #185

Problem

A closed-set member is its C# field's name everywhere: parsing, JSON, the EF Core column, the OpenAPI enum and the fault identifiers. So renaming a field changes stored data and the wire contract, and a closed set can't be stored or sent as a number or a fixed key. Ardalis.SmartEnum, Intellenum, Thinktecture.Runtime.Extensions and EricksonLopez.DomainPrimitives all allow a value apart from the identifier (docs/comparison.md, "Choose something else when"). The roadmap lists this under "2.x, with a spike first" (proposed in the review of 2026-10-13).

Change

A spike first, under spikes/, with a README giving the question, the answer and the evidence. It decides:

• How a member declares its external name. It must be declared, not inferred, and must survive Native AOT. Options are a constructor argument, an attribute on the field, or a key type on the set's interface.
• Whether a number is in scope, or only a string apart from the identifier.
• What follows the name at once: Parse/TryParse (and ignoreCase), JSON, the EF Core column and its translation, the OpenAPI enum, and fault identifiers. A partial rollout would make the name different in different places.
• The refusals. Two members with one external name, or with names that differ only in case under ignoreCase, need a generator error like CMTK1021. An Error is allowed in a minor because only the new API can write that form (decision 118).
• What existing sets do. A set that declares no external name keeps today's behaviour byte for byte.

Then either the feature lands in 2.4.0 under a decision record, or the spike's answer moves it back to the roadmap with the reason.

Done when

• The spike's README answers the questions above, and a decision record states the choice.
• If it lands:
  • Parsing, JSON, EF Core, OpenAPI and faults all use the external name, each with a test.
  • The refusals are tested.
  • The closed-set docs, the Generators README, the comparison's closed-set table and the changelog say so.
  • Claims in test/oracles/claims.json hold it.