Java: index Lombok-generated members, so calls to them resolve by type #2662

What problem does this solve?

Lombok writes getters, setters, builders and withers at compile time, so the source has no declaration for invoice.getNumber() or Invoice.builder().number("1").build(). The graph has no node for them, the type-aware resolver cannot find them, and the call binds by name to any project method that happens to share it.

// src/com/example/model/Invoice.java
@Data @Builder
public class Invoice { private String number; private boolean paid; }

// src/com/example/model/Report.java: an unrelated class
public class Report {
    public String getNumber() { return "r"; }
    public static Report builder() { return new Report(); }
}

// src/com/example/app/Caller.java
public String number(Invoice invoice) { return invoice.getNumber(); }
public Invoice make() { return Invoice.builder().number("1").paid(true).build(); }

On main (v0.12.0): number → Report.getNumber (unique_name); make → the fields Invoice.number and Invoice.paid (suffix_match, unique_name); builder() and build() get no edge. On a real Spring service with Lombok, about 8,300 generated members are missing, and Dto.builder() calls bind to whichever project class has the first builder by name.

Proposed solution

In extraction, for a class or field with a Lombok annotation, add the members Lombok would generate as Method (and builder Class) definitions, marked generated and tagged "lombok" in their properties:

• only annotations that the file's imports make Lombok's (import lombok.Data;, import lombok.*;, or @lombok.Data); Spring's @Value generates nothing;
• a member the source declares by name is not generated, as Lombok itself skips it;
• @Getter, @Setter, @Data, @Value (fields final unless @NonFinal), @With, AccessLevel.NONE; then @Builder/@SuperBuilder (also on records, @Builder.Default, builder options) and @Accessors(fluent, chain, prefix);
• registered by QN only: generated members never enter the by-name index, so the name guess can never pick one; only the type-aware resolver reaches them. On the measured service, registering them by name as well let the name guess land 14,061 extra suffix_match edges on them.

Cost on that service, measured for this together with two smaller resolver fixes: index time about +0.5 s (+10 %), DB +52 MB (+21 %), mostly the generated nodes and their edges.

I have this implemented in two parts (accessors first, then builders and @Accessors), each with tests, and would like a maintainer view before opening the PRs: is a generated member in the definition list acceptable, and is QN-only registration the right way to keep it out of the name guess?