Foundation Classes, Convert - Replace handle-based APIs with direct array access (#1057)

Refactor the Convert package to eliminate heap-allocated handle-based storage
in favor of direct NCollection_Array members, improving performance and
simplifying the API. Deprecate single-element accessors (Pole, Knot, etc.)
in favor of batch const-reference accessors (Poles, Knots, etc.).

Convert_ConicToBSplineCurve:
- Replace handle members (poles, weights, knots, mults) with direct
  NCollection_Array1 fields (myPoles, myWeights, myKnots, myMults).
- Replace BuildCosAndSin handle-based parameters with array references.
- Add batch accessors: Poles(), Weights(), Knots(), Multiplicities().
- Deprecate single-element accessors: Pole(), Weight(), Knot(), Multiplicity().
- Update all conic subclasses: Circle, Ellipse, Hyperbola, Parabola.

Convert_ElementarySurfaceToBSplineSurface:
- Replace handle members with direct NCollection_Array fields
  (myPoles, myWeights, myUKnots, myVKnots, myUMults, myVMults).
- Add Finalize() to trim oversized arrays in derived constructors.
- Add batch accessors: Poles(), Weights(), UKnots(), VKnots(),
  UMultiplicities(), VMultiplicities().
- Deprecate single-element accessors: Pole(), Weight(), UKnot(), VKnot(),
  UMultiplicity(), VMultiplicity().
- Update all surface subclasses: Cone, Cylinder, Sphere, Torus.

Convert_CompPolynomialToPoles / Convert_GridPolynomialToPoles:
- Replace handle-based output parameters with direct const-reference
  accessors for Poles, Knots, Multiplicities.
- Deprecate old handle-based Poles(), Knots(), Multiplicities() overloads.

Convert_CompBezierCurvesToBSplineCurve (2D and 3D):
- Extract common logic into Convert_CompBezierCurvesToBSplineCurveBase
  template header to eliminate code duplication.
- Replace handle<HArray1> members with direct NCollection_Array1 storage
  in the internal sequence, removing unnecessary heap indirection.

NCollection_Sequence:
- Fix Node constructors to use member initializer lists (copy/move
  construction) instead of default-construct + assign, which failed for
  types like NCollection_Array1 where operator= requires matching sizes.

Downstream callers migrated:
- AdvApprox_ApproxAFunction: use new const-ref Knots()/Multiplicities().
- AppDef_Variational: use new const-ref Knots()/Multiplicities().
- AdvApp2Var_ApproxAFunc2Var, AdvApp2Var_Patch: use new const-ref API.
- Geom2dConvert, GeomConvert, GeomConvert_1: use new const-ref API.
- GeomFill_PolynomialConvertor, GeomFill_QuasiAngularConvertor: adapted.
- Geom_OsculatingSurface: use direct array references instead of
  handle->Array*() calls.

Added GTests for all Convert classes covering conic curves,
elementary surfaces, CompBezier, CompPolynomial, and GridPolynomial
conversions.

src/ModelingAlgorithms/TKGeomAlgo/GeomFill/GeomFill_PolynomialConvertor.cxx

@@ -20,8 +20,8 @@
 #include <gp_Pnt.hxx>
 #include <gp_Vec.hxx>
 #include <PLib.hxx>
-#include <StdFail_NotDone.hxx>
 #include <Standard_Integer.hxx>
+#include <StdFail_NotDone.hxx>
 #include <NCollection_Array1.hxx>
 #include <NCollection_HArray1.hxx>
 #include <NCollection_Array2.hxx>
@@ -77,7 +77,7 @@ void GeomFill_PolynomialConvertor::Init()
               Coeffs,
               Inter,
               TrueInter); En attente du bon Geomlite*/
-  AConverter.Poles(Poles1d);
+  Poles1d = new NCollection_HArray2<double>(AConverter.Poles());
 
   for (jj = 1; jj <= Ordre; jj++)
   {

src/ModelingAlgorithms/TKGeomAlgo/GeomFill/GeomFill_QuasiAngularConvertor.cxx

@@ -93,7 +93,7 @@ void GeomFill_QuasiAngularConvertor::Init()
 
   // Convertion
   Convert_CompPolynomialToPoles AConverter(Ordre, Ordre - 1, Ordre - 1, Coeffs, Inter, TrueInter);
-  AConverter.Poles(Poles1d);
+  Poles1d = new NCollection_HArray2<double>(AConverter.Poles());
 
   for (jj = 1; jj <= Ordre; jj++)
   {