fix(#916): reset OCCTSectionBuilder's built flag on failed rebuild

[gsdali]

What & why

OCCTSectionBuilder's built field (gating ancestorFaceOn1(edge:)/ancestorFaceOn2(edge:)) was
only ever set true on a successful Build() and never reset on a failed rebuild — the same
staleness class PR #912 fixed for the sibling OCCTThruSections struct (#910). A SectionBuilder
reused via init1/init2 + build() that had already built successfully once kept built == true through a later failed rebuild, so ancestorFaceOn1/ancestorFaceOn2 skipped their guard
and read state from the failed/incomplete rebuild's BRepAlgoAPI_Section instance.

No live repro existed when #916 was filed — a quick attempt (two far-apart, non-intersecting
boxes) just produces an empty, still-IsDone() section, not a failure. This PR finds one. Read
BRepAlgoAPI_BooleanOperation::Build()/BOPAlgo_PaveFiller::Init() directly
(Libraries/occt-src/...) and swept 13 candidate triggers against the real pinned kernel:
self-intersecting/bowtie faces, coincident/duplicate solids (including two exactly-coincident
spheres), an empty compound, a degenerate collinear-point face, NaN and zero-coefficient plane
coefficients, an invalid #905-style uncapped loft solid, and the too-few-arguments path. None make
Build() cleanly fail on a builder that already built successfully once. The only clean
(non-throwing) trigger found is a literal null TopoDS_Shape argument
(BOPAlgo_PaveFiller::Init()'s own AddError(new BOPAlgo_AlertNullInputShapes) check) — and
Shape never wraps one, so it is not reachable through OCCTSwift's public Swift API.

Proved the defect live anyway, at the bridge boundary rather than through Swift: compiled the real,
unmodified OCCTBridge_Modeling.mm and drove its actual exported OCCTSectionBuilder* C
functions with a hand-constructed null-wrapping OCCTShape as the one input Swift's type system
can't produce. This found something worse than the issue anticipated — an uncatchable SIGSEGV,
not just a stale answer: the failed rebuild's BOPAlgo_PaveFiller::Init() bails via
AddError(BOPAlgo_AlertNullInputShapes) before setting its own myDS, and
HasAncestorFaceOn1/HasAncestorFaceOn2 — reached only because built was wrongly still true —
dereferences it. Full writeup, driver, and before/after transcripts:
Scripts/repro/916-sectionbuilder-built-flag-stale/.

The fix mirrors #912's OCCTThruSections pattern: built = false on both the !IsDone() and
catch (...) paths in OCCTSectionBuilderBuild. It also mirrors #912's own second-review finding
(the AddWire/AddVertex staleness gap): all six
Init1Shape/Init1Plane/Init1Surface/Init2Shape/Init2Plane/Init2Surface functions now
reset built = false on success too, closing the adjacent "re-init after a successful build,
without rebuilding" gap — which is reachable through pure Swift (unlike the null-shape SIGSEGV
above), and is what the new Swift regression test exercises.

Closes #916

CHANGELOG entry

SectionBuilder no longer returns stale results after a failed rebuild (#916)

ancestorFaceOn1(edge:) and ancestorFaceOn2(edge:) read post-build OCCT state gated on a
bridge-side built flag that was only ever set true, never reset on a failed rebuild. A
SectionBuilder reused across multiple build() calls could silently keep answering from a prior
successful build's geometry after a later build() call on the same instance genuinely failed —
and, in the worst case (an argument that fails BRepAlgoAPI_Section::Build()'s own internal
null-shape check), could crash instead of returning nil. Calling init1/init2 again after a
successful build without a following build() call had the identical staleness gap. Fixed by
resetting the tracked build outcome on every path that invalidates the previous result.

SemVer impact

PATCH. A caller of SectionBuilder.ancestorFaceOn1(edge:)/.ancestorFaceOn2(edge:) who reuses a
builder across build() calls, or calls init1/init2 again without a following build(), now
correctly gets nil instead of stale geometry from an earlier build. No signature change; the only
observable difference is a previously-buggy return value becoming correct (or, in the crash case,
no longer crashing). No migration.

Checklist

• New or changed behavior is covered by a unit test in the same PR (not just manual
  verification): ancestorFaceNilAfterReinitWithoutRebuild in
  Tests/OCCTStressTests/StressBuilderLifecycleTests.swift, covering the half of OCCTSectionBuilder's built flag never resets to false on a failed rebuild (same class as #910) #916 reachable
  through the public Swift API (the Init*-without-rebuild staleness gap). The other half (a
  genuinely failed build() itself, via the null-shape mechanism) is proven at the bridge
  boundary instead — see Scripts/repro/916-sectionbuilder-built-flag-stale/ — since it isn't
  reachable through Swift's type-safe Shape API. Also fixed a fixture bug while writing this
  test: Shape.box(width:height:depth:) (no origin) centers at origin, while
  Shape.box(origin:width:height:depth:) treats origin as a corner — mixing the two (as an
  early draft of this test did, copying SectionBuilderTests.sectionTwoShapes's own fixture
  verbatim) produces boxes that only touch tangentially along a shared face plane, not a genuine
  3D overlap, so no section edge ever resolved an ancestor face. Verified directly in C++ against
  the real kernel before rewriting the fixture.
• Every new test and every new --self-test case was run once with its subject broken, and the
  failure is reported here: reverted the built = false reset in OCCTSectionBuilderInit1Shape
  only, re-ran ancestorFaceNilAfterReinitWithoutRebuild — it failed with both
  ancestorFaceOn1(edge:) and ancestorFaceOn2(edge:) wrongly returning non-nil (Expectation failed: ... == nil on both). Restored the fix; the test and the rest of
  StressSectionBuilderLifecycleTests (7/7) pass again. Separately, at the bridge level: ran
  occt_916_section_builder_stale against the unmodified bridge file first (SIGSEGV, see
  before.txt), then against the fixed one (clean exit, correct nil, see after.txt).
• The CHANGELOG entry above is complete, and docs/CHANGELOG.md is not in this diff.
• The SemVer impact above is stated, and docs/SEMVER.md is not in this diff.

Notes for the reviewer

code-style is expected red here, for the same reason as #912: OCCTBridge_Modeling.mm is
still on Scripts/style-manifest-bridge.txt (a full clang-format pass is ~24,000 diff lines for
this file alone — disproportionate for a 2-function-signature change, and tracked by #917, a
sub-issue of #876). code-style is not a required status check (only gate-scripts is, confirmed
via the repo's rulesets), so this doesn't block merge. Unlike #912, this PR did bring
SectionBuilder.swift into full swift-format/swiftlint compliance and remove it from
Scripts/style-manifest-swift.txt — matching #912's own precedent for the analogous small Swift
file (import order, one blank-comment-line-after-summary fix per doc comment, one line wrap).

Verified: StressSectionBuilderLifecycleTests (7/7), full OCCTModelingTests (645/645), full
OCCTStressTests (361/361), all six static gate scripts (check-bridge-index,
check-null-handle-guards, check-docs-defaults, check-docs-existence,
derive-bridge-header-split --verify, count-operations), and check-style-manifest.py --base origin/main (clean except the expected, precedented OCCTBridge_Modeling.mm exemption above).

[gsdali]

Self-review pass, per request, before your own review. Fixed the code-style failure and independently verified (not just re-read) every substantive claim in the PR body.

code-style fix: merged current origin/main (brings in #912's already-merged # deferred, see #917 annotation on OCCTBridge_Modeling.mm in style-manifest-bridge.txt, plus resolved a real, small conflict in style-manifest-swift.txt — this branch and an unrelated already-merged PR each removed a different file from the same list). Updated the annotation to note this is now confirmed recurring (#912 and #918 both hit it) rather than a one-off. code-style still fails, for the same tracked/deferred reason as #912/#915 — not a new problem.

Self-review findings:

• Re-ran the SIGSEGV repro myself rather than trusting before.txt/after.txt: compiled the unmodified bridge, reverted the fix via git apply -R, reproduced exit 139 directly. Caught my own mistake along the way — a first attempt reverted only OCCTSectionBuilderBuild and got a false negative, because the driver's Init1Shape call still carried the other half of the fix and independently reset built before the trigger could fire. Redid it as a full reverse-patch.
• Re-ran the "prove the test fails" claim for `ancestorFaceNilAfterReinitWithoutRebuild" by injecting the same defect myself — matches the PR body exactly.
• Verified the Shape.box(origin:...) corner-vs-center fixture claim against the actual doc comments.
• The one question this PR's own history should have raised but didn't address: does OCCTSectionBuilder have fix(#910): guard ThruSectionsBuilder.generatedFace(from:) on IsDone() #912's finding-1 class of gap — a built flag insufficient on its own because internal state accumulates across builds (needing a membership check too, the way OCCTThruSections ended up needing)? Checked the actual OCCT source (BRepAlgoAPI_BuilderAlgo::IntersectShapes): myDSFiller is delete+new'd fresh on every Build() call — architecturally the opposite of ThruSections' additive-only myEdgeFace map. A simple built flag is genuinely sufficient here; confirmed empirically with a throwaway probe that setApproximation/computePCurveOn1/2 not resetting built is correct, not a gap (SectionBuilder has no cached .shape accessor to go stale — build() always re-executes fresh, unlike ThruSectionsBuilder).
• Found and fixed one real gap: no fenced ```swift snippet on the changed public API (CLAUDE.md's Documentation Standards — the same class of finding fix(#910): guard ThruSectionsBuilder.generatedFace(from:) on IsDone() #912's own round 2 caught). Added to ancestorFaceOn1(edge:)/`ancestorFaceOn2(edge:)`, plus `- Parameter`/`- Returns` structure.

All 5 required static gates clean, full swift test (5611/5611), SectionBuilder-specific suites (11/11) re-verified after every change. Pushed as fac32a1.