#2932 Use builder for generic target types - #4133

Fixes #2932

Root cause

DefaultBuilderProvider#isBuildMethod checks that the builder's build method returns the type being built via typeUtils.isAssignable( buildMethodReturnType, typeElement.asType() ). For a generic type such as Lombok's @Builder on Record<T>, the build method returns Record<T> where T is the builder's type variable, while typeElement.asType() is Record<T> with the class's own T. These are distinct type variables, so the check fails for every generic type, no builder is found, and MapStruct falls back to the (private) all-args constructor — exactly the behaviour reported in the issue.

Change

• DefaultBuilderProvider#isBuildMethod: compare the erasures of the build method's return type and the type being built.
• BuilderType#create: resolve the builder's type parameters against the concrete target type, so that for Record<String> the generated code uses RecordBuilder<String> (not a raw/unresolved RecordBuilder<T>). If the parameters can't be resolved, the builder type is left unchanged.

Tests

Added processor/src/test/java/org/mapstruct/ap/test/bugs/_2932/ covering both builder shapes:

• shouldUseBuilderOfGenericTarget — Record<T> with a static builder() factory (mirrors what Lombok generates for @Builder on a generic class).
• shouldUseConstructorCreatedBuilderOfGenericTarget — Pair<K, V> whose builder is created through its public constructor.

Both fail on current main (4/4 failures across the JDK and Eclipse compilers) and pass with the fix.

Verification done:

1. No in-flight PR or linked branch for Builder not Invoked for Generic Classes #2932 (gh pr list --search, issue has no linked PRs).
2. No active claim on the issue (unassigned, no comments).
3. Code-focused fix (.java only).
4. Reproduced on current main (211b2be) by running the new test against the pre-fix code — all 4 iterations fail; they pass after the fix.
5. No related open epic/tracking issue found.
6. ./mvnw -pl processor -am clean install: all 3559 processor tests pass (0 failures, 0 errors) and checkstyle passes. (Locally on JDK 25 the animal-sniffer check reports TypeKind.MODULE in the untouched Type.java; this is unrelated to this change.)