OCCTSectionBuilder's built flag never resets to false on a failed rebuild (same class as #910)

[gsdali]

OCCTSwift version

Current main (found during PR #912's review, unreleased).

Platform

macOS (arm64), likely all platforms — bridge code, not platform-specific.

Description

OCCTSectionBuilder (Sources/OCCTBridge/src/OCCTBridge_Modeling.mm, backs SectionBuilder) has
the identical staleness defect PR #912 just fixed for OCCTThruSections, in a sibling reused
builder in the same file:

struct OCCTSectionBuilder {
    BRepAlgoAPI_Section section;
    bool built;
    OCCTSectionBuilder() : section(), built(false) {}
    OCCTSectionBuilder(const TopoDS_Shape& s1, const TopoDS_Shape& s2) : section(s1, s2, false), built(false) {}
};

OCCTShapeRef OCCTSectionBuilderBuild(OCCTSectionBuilderRef builder) {
    if (!builder) return nullptr;
    try {
        builder->section.Build();
        if (!builder->section.IsDone()) return nullptr;
        builder->built = true;
        return new OCCTShape{builder->section.Shape()};
    } catch (...) { return nullptr; }
}

OCCTShapeRef OCCTSectionBuilderAncestorFaceOn1(OCCTSectionBuilderRef builder, OCCTShapeRef edge) {
    if (!builder || !edge || !builder->built) return nullptr;
    ...
}

built is only ever set to true, on a successful build. On a failure — !IsDone() — the
function returns nullptr without ever resetting built back to false. So on a SectionBuilder
reused via init1(...)/init2(...) + build() (matching the reuse pattern PR #912's own
regression tests exercise for ThruSectionsBuilder), a builder that already built successfully
once keeps built == true through a later failed rebuild — ancestorFaceOn1(edge:) and
ancestorFaceOn2(edge:) would then skip their guard and read state from the failed/incomplete
rebuild's BRepAlgoAPI_Section instance instead of correctly returning nil.

Steps to reproduce

Not yet reduced to a minimal repro that reliably makes BRepAlgoAPI_Section::Build() itself fail
(IsDone() == false) — a quick attempt using non-intersecting/far-apart shapes did not trigger a
failure (the section is just empty, still IsDone() == true). Whoever picks this up needs to find
an actual failure trigger for BRepAlgoAPI_Section::Build() first (an invalid plane/surface
argument via init1(plane:)/init2(surface:) is the likely candidate, unexplored) before this can
be proven live the way PR #912's own regression tests were.

Notes

Found during PR #912's review (the #910 fix) while confirming that built is an established
naming convention in this bridge (it is — this struct predates #912's own use of the same name for
OCCTThruSections). Filed separately per this project's scope-boundary policy rather than folded
into #912, and because it needs its own live repro before a fix can be validated the same way #910
was.

Candidate fix, mirroring #912's own pattern: builder->built = false; on the !IsDone() path (and
on the catch (...) path) in OCCTSectionBuilderBuild, before returning nullptr.

[github-actions]

This issue is missing a priority:* label (priority:P0–priority:P3). Please add the appropriate label so it shows up correctly in triage.