BUG: Remove shadow declarations of NumberOfClasses in RGBGibbsPriorFilter

[hjmjohnson]

Remove RGBGibbsPriorFilter's redeclarations of m_NumberOfClasses and m_MaximumNumberOfIterations that shadowed the inherited MRFImageFilter members, along with the 4 hand-written accessor overrides that existed only to redirect at the shadow. New GTest suite confirms zero observable API change (defaults, Set/Get via subclass and base pointer, MTime semantics).

The shadow and why it was wrong

itkRGBGibbsPriorFilter.h declared:

unsigned int m_NumberOfClasses{ 0 };        // shadows MRFImageFilter::m_NumberOfClasses (also 0)
unsigned int m_MaximumNumberOfIterations{ 10 };  // shadows MRFImageFilter::m_MaximumNumberOfIterations (default 50)

And four hand-written accessor overrides whose only purpose was to direct this->m_* at the subclass shadow rather than the inherited base member:

void SetNumberOfClasses(const unsigned int numberOfClasses) override { ... }
unsigned int GetNumberOfClasses() const override { ... }
void SetMaximumNumberOfIterations(const unsigned int numberOfIterations) override { ... }
unsigned int GetMaximumNumberOfIterations() const override { ... }

Virtual dispatch through the base pointer made the shadow work correctly for Set/Get callers, but the design carried real costs:

• Every instance held two orphaned 8-byte fields (the inherited base members were never read or written, because RGBGibbsPriorFilter::GenerateData() is a complete override that never calls Superclass::GenerateData()).
• Any future MRFImageFilter method that read m_NumberOfClasses or m_MaximumNumberOfIterations directly would silently see stale defaults for RGBGibbsPriorFilter instances — a regression hazard hidden by virtual dispatch.
• m_NumberOfClasses{0} in both classes meant one of the shadows was purely accidental — only the MaximumNumberOfIterations shadow served the design goal of changing the default from 50 to 10.

The fix

1. Delete the 4 redundant accessor overrides in itkRGBGibbsPriorFilter.h. The inherited MRFImageFilter::SetNumberOfClasses and friends (declared via itkVirtualSetMacro / itkGetConstMacro) now operate on the single inherited member.
2. Delete the shadow member declarations in itkRGBGibbsPriorFilter.h.
3. Preserve the subclass-specific default MaximumNumberOfIterations = 10 (vs base default 50) by calling the inherited setter from the RGBGibbsPriorFilter constructor: this->SetMaximumNumberOfIterations(10);. m_NumberOfClasses default 0 matches the base, so no constructor work is needed.
4. Drop 2 redundant PrintSelf lines that re-printed NumberOfClasses and MaximumNumberOfIterations already emitted by Superclass::PrintSelf.
5. Switch the one direct-field reference in .hxx (m_ClassifierPtr->SetNumberOfClasses(m_NumberOfClasses);) to this->GetNumberOfClasses() — the inherited member is private in the base, so the public accessor is the right entry point.

Net diff: −43 lines in the header/hxx, +2 lines.

Test methodology — characterize-then-refactor

A new itkRGBGibbsPriorFilterGTest.cxx was added before the fix and committed first, so the test suite establishes the observable API contract independent of the code change. The fix commit is then verified non-regressive against that suite.

The 9 GTests cover:

• RGBGibbsPriorFilter.DefaultNumberOfClassesIsZero — default 0
• RGBGibbsPriorFilter.DefaultMaximumNumberOfIterationsIsTen — default 10 (the subclass-specific value)
• MRFImageFilter.DefaultMaximumNumberOfIterationsIsFifty — base default 50 (documents the contract difference)
• RGBGibbsPriorFilter.SetGetNumberOfClassesViaSubclass
• RGBGibbsPriorFilter.SetGetMaximumNumberOfIterationsViaSubclass
• RGBGibbsPriorFilter.SetGetNumberOfClassesViaBasePointer — verifies virtual dispatch through MRFImageFilter*
• RGBGibbsPriorFilter.SetGetMaximumNumberOfIterationsViaBasePointer — same
• RGBGibbsPriorFilter.ModifiedTimeAdvancesOnNumberOfClassesChange — MTime advances when the value changes
• RGBGibbsPriorFilter.ModifiedTimeUnchangedOnIdenticalAssignment — MTime unchanged on no-op assignment

All 9 pass on the pre-fix tree (commit 1) and the post-fix tree (commit 2). The pre-existing itkRGBGibbsPriorFilterTest image-pipeline CTest also continues to pass — RGBGibbsPriorFilter::GenerateData() now operates on the inherited fields end-to-end.

Local verification

• pre-commit run --all-files clean on both commits.
• Full ITK ninja build: 0 real errors (the only failures are pre-existing KWStyle ExternalProject SDK issues unrelated to this PR).
• ctest -R "MarkovRandomFields|RGBGibbs|MRFImageFilter": 13/13 pass (9 new GTests + 2 framework + 2 existing CTests).
• Broader ctest -R Segmentation: all pass except itkVectorThresholdSegmentationLevelSetImageFilter, which is in Modules/Segmentation/LevelSets/ and has zero file-level dependency on either of the modified files (verified via grep -rln "MRFImageFilter\\|RGBGibbsPriorFilter" Modules/Segmentation/LevelSets/ → no matches). Pre-existing baseline-PNG drift unrelated to this PR.

[hjmjohnson]

@greptile review

[dzenanz]

Looks good on a glance.