Shape Healing - create a ReShape context when none is set in ShapeFix_ComposeShell and ShapeUpgrade_WireDivide

[gsdali]

Fixes #1409.

Problem

ShapeFix_ComposeShell::Perform(), ShapeFix_ComposeShell::SplitEdges() and ShapeUpgrade_WireDivide::Perform() dereference Context() unconditionally. ShapeFix_Root::Context() returns myContext, which the base constructor leaves null and only an explicit, optional SetContext() ever fills — so Init(...) + Perform(), the usage both classes' public API invites, is a null-handle dereference (SIGSEGV at address 0, uncatchable in-process). A plain 4-edge planar square face reproduces it 100% of the time; see #1409 for a self-contained reproducer.

The reason this has gone unnoticed: in-kernel, both classes are only ever driven by ShapeUpgrade_FaceDivide::Perform(), which already self-creates a context and hands it down (ShapeUpgrade_FaceDivide.cxx:185 for the compose shell, :238 for the wire divide). Reached that way neither class ever sees a null context, so the existing tests never exercise the crash.

Fix

Add the guard that nine other classes in the same package already use — ShapeFix_Shape::Init, ShapeFix_Shell::Perform, ShapeFix_Solid::Perform, ShapeFix_FixSmallFace::Init, ShapeFix_SplitCommonVertex::Init, ShapeFix_Wireframe (both entry points), ShapeFix_Wire::FixGap3d/FixGap2d, ShapeFix_Face::FixMissingSeam, ShapeUpgrade_ShapeDivide::Perform — to the three public entry points that were missing it:

if (Context().IsNull())
{
  SetContext(new ShapeBuild_ReShape);
}

ShapeFix_ComposeShell's other context-dereferencing methods (LoadWires, SplitWire, SplitByLine, MakeFacesOnPatch, DispatchWires) are all reached through Perform() or SplitEdges(), so guarding those two covers them; they are also const, so they could not create a context themselves.

Same defect class as #1378 / #1380 (merged), which fixed ShapeFix_Face::FixPeriodicDegenerated.

Validation

Verified downstream on V8_0_0_p1 with the two patched translation units linked ahead of the stock archive, so only these symbols change:

• Crash closed — the reproducer from Shape Healing - ShapeFix_ComposeShell::Perform/SplitEdges and ShapeUpgrade_WireDivide::Perform SIGSEGV on a null ReShape context #1409, plus the same two calls against an unbounded-cylindrical face, SIGSEGV 100% of the time before and complete normally after.
• No behaviour change on the working path — with a context set (the only path that worked before), the result is byte-identical before and after: BRepTools::Write dump hash plus face/wire/edge/vertex counts, for both a planar and a cylindrical face, both classes. The guard only fires when there was no context, and in that case it produces exactly what an explicitly-provided context produces.
• clang-format clean on both files.

[dpasukhi]

Dear @gsdali please add GTest to cover your changes, you can follow the logic of all exited GTests.
Previously you already added them.
Also please clean up code with too long comments

[gsdali]

Thanks — pushed an update.

GTests added: ShapeFix_ComposeShell_Test.cxx and ShapeUpgrade_WireDivide_Test.cxx in src/ModelingAlgorithms/TKShHealing/GTests/, following the existing ShapeUpgrade_FaceDivide_Test.cxx's pattern for this exact defect class (issue #1179). Each covers the entry point that was crashing (Perform() on both classes, plus SplitEdges() on ShapeFix_ComposeShell) with no context set, and a with/without-context equivalence check. Verified both directions before pushing: linked against a build with the guard reverted, all three *_WithoutContext_DoesNotCrash cases SIGSEGV (exit 139); linked against the guarded build, all five tests pass.

Comments shortened: dropped the three explanatory comments on the new guards entirely, matching the nine sibling guards already in this file/package (e.g. ShapeFix_Shape::Init), none of which carry one.