[CaffeinatedCoder]

Closes #226

Stacked on #218 (fix/closed-set-base-member-refused-2-3-1), whose base-class reads for closed sets this fix's closed-set rows need. Merge #218 first; this PR then retargets to release/2.3.1.

Problem

ValueObjectDeclaration.MethodsWithSignature and ClosedSetTypes.SameParameters matched a hand-written method to a generated one by name and parameter types only. A legal overload was refused with CMTK1011 as the generated member written by hand: Parse<TFormat>(string, IFormatProvider?), From(in int), From(ref int), a TryParse whose last parameter is by value, a generic or by-value TryFormat, and a closed set's generic Parse or by-value TryParse. That happened on the type, on a base record and on a set's base class. C# overloads by each of these, and Metalama introduces the generated member beside it (measured in a scratch consumer: every row builds and calls the generated member).

The same catalogue, run over the other name-and-parameters checks, found seven siblings (Found-By: VA). Three were silent:

• A generic ToString<TFormat>() was taken for the seam. The value object printed ToStringOverloadedByArity { Value = a }, as a record does, and a closed set printed its type name.
• A generic or by-in CompareTo(TSelf) was taken for the ordering seam, so the ordering was marked [HandWrittenOrdering] and EF Core refused to order by it.
• A generic Equals<T>(TBase?) stopped the value object re-implementing a deeper base's explicit IEquatable<TBase>, so a HashSet<TBase> held two value objects holding 42.

The other four were refusals: a generic or by-reference CompareTo(object), TryFrom, Revalidate, Equals, GetHashCode or PrintMembers, on the type or a base record.

Change

• MethodsWithSignature and SameParameters match only a method without type parameters, and compare each parameter's ref kind: by reference where the generated one is out, by value where it's by value. ref or in against out (CS0663) and nullability stay CMTK1011.
• ValueObjectDeclaration.IsByValueAndNotGeneric holds the seam and synthesized-member checks to the same rule: IsHandWrittenToString, InheritedSeams, PrintMembersDeclaredByHand, EqualityDeclaredByHand, InheritedEquality, the comparable aspect's HandWrittenComparisonMembers and DeclaresCompareTo, and the contract aspect's re-implemented IEquatable check.
• Generators README ("Write your own members", "Rules for a closed set"), design/aspects.md, three claims, two CHANGELOG lines.
• Test infrastructure: FabricTests looks up From/TryFrom by signature, since an overload made it ambiguous. The surface snapshot now prints a generic method's type parameters and in/ref, so an overload doesn't read as a duplicate line. Match gains <TResult>.

Tests

• OverloadTests (new): 53 rows where the generated member sits beside a hand-written overload and is what a plain call reaches, 14 rows that reach the hand-written one, and 3 that check the ordering isn't marked hand-written. They cover the type (record struct, sealed record class), a base record in this assembly and in another (ProbeOverloadingRecord), two levels up, a closed set and its base class.
• BuildOutcomeTests: 8 control rows stay refused: ref/in against out on the type, a base record, a set and a set's base, a ref TryFormat, and another nullability.
• Reverted: without the first commit, all 16 new types are refused (32 CMTK1011). Comparing ref kinds exactly fails the 6 ref/in controls. Without the second commit, 5 of its types are refused. Reverting IsHandWrittenToString, DeclaresCompareTo or the contract aspect's check alone fails its own rows (3, 2 and 1).
• Locally through tools/BuildVerdict: Generators.Tests 1249 passed, Conventions.Tests 2016, Samples 150.

🤖 Generated with Claude Code