ENH: Enforce computational requirements for SobelOperator

* Sobel Direction 1 coefficients (G_y) convolved with Neighborhood of 1's
  should equal 0

         G_y           *      N
  [ +1.   +2.   +1. ]    [ 1  1  1]
  [  0.    0.    0. ]  * [ 1  1  1]    =>   0
  [ -1.   -2.   -1. ]    [ 1  1  1]

If the SobelOperator is templated over a pixel type
that is not signed, the G_y coefficients above are
typecast to the internal pixel type, this removing
the sign, or introducing wrap errors.

Test code:
    double d = -1.;
    unsigned char c = d;
    std::cout << +c << std::endl;

On intel computers:
    prints 255
On apple silicon computers:
    print 0

NOTE: The sobel coeficients are not properly applied
      during the inner product for unsigned data types.

Turn on concept checking for the SobelOperator to enforce
signed neighborhood processing.

Modules/Core/Common/include/itkSobelOperator.h

@@ -121,6 +121,12 @@ class ITK_TEMPLATE_EXPORT SobelOperator : public NeighborhoodOperator<TPixel, VD
   // virtual void CreateToRadius(const unsigned long);
 
 protected:
+#ifdef ITK_USE_CONCEPT_CHECKING
+  // Begin concept checking
+  itkConceptMacro(SignedOutputPixelType, (Concept::Signed<typename NumericTraits<TPixel>::ValueType>));
+  // End concept checking
+#endif
+
   /** Type alias support for coefficient vector type.*/
   using typename Superclass::CoefficientVector;
 

Modules/Core/Common/include/itkSobelOperator.hxx

@@ -58,7 +58,10 @@ SobelOperator<TPixel, VDimension, TAllocator>::Fill(const CoefficientVector & co
       for (int x = -1; x <= 1; ++x)
       {
         const int pos = center + y * this->GetStride(1) + x * this->GetStride(0);
-        //
+        // Note, The following line copies the double precision
+        // coefficients of SobelOperator to the pixel type
+        // of the neighborhood operator which may not support
+        // negative numbers, or floating point numbers.
         this->operator[](pos) = static_cast<TPixel>(coeff[coeff_index]);
 
         ++coeff_index;