[hjmjohnson]

Fixes two image iterator bugs on main and adds tests for the iterators' current write access, conversions, and copies. This splits the fixes and the coverage that does not depend on the rework out of #7065, which will be rebased on this PR.

Requires #7068 to merge first: its two commits come first on this branch; review from "BUG: Keep the position and reflection state in reflective iterator copies".

• BUG: Copying a ReflectiveImageRegionConstIterator rewound it, and operator= dropped m_IsFirstPass.
• BUG: PathIterator::Value() did not compile: it called a GetImage() member and read an m_ImageIndex member that do not exist.
• ENH: Tests for Set() and Value() access, one-way conversion from writable to read-only iterators and their comparison, deduction from raw image pointers, reverse traversal of a sub-region, and copies that outlive their source.

Local verification

• macOS, AppleClang, ITK build tree: ITKCommon, ITKDistanceMap, ITKImageFrequency, and ITKPath test drivers build with 0 warnings; 64 of 64 iterator, DistanceMap, and Danielsson tests pass.
• The same test files compiled against ENH: Select image iterator access by image constness (supersedes #7021) #7065's tip also build with 0 warnings and pass 64 of 64. The one change needed there is ENH: Select image iterator access by image constness (supersedes #7021) #7065's own edit of ReverseIteratorConversionCallsNonConstGetBufferPointer.
• Before the fix, ReflectiveImageRegionIterator.ConversionToConstIteratorKeepsPosition fails; RandomNonRepeatingConversionOutlivesSource depends on BUG: Copy the shuffled order when assigning a RandomPermutation #7068's copy constructor.
• pre-commit run --all-files passes.

Left for #7065

These checks do not hold on main and stay with the rework: ImageReverseConstIterator and ImageRegionReverseConstIterator provide Set() through a const_cast; LineConstIterator, PathConstIterator, and ReflectiveImageRegionConstIterator have no deduction guide; only ImageIteratorWithIndex has ImageIteratorDimension.

[hjmjohnson]

Force-push 8a2fce5e10d..9229cb6c578 changes content only, with the same base. It fixes the Pixi-Cxx windows-2022 failure: MSVC error C2039 in itkFloodFillIteratorGTest, where the HasSet checks paired the ImageFunction flood-fill iterators with SphereSpatialFunction, which has no EvaluateAtIndex().

What changed (one commit: "ENH: Test image iterator write access, conversions, and copies")

• The ImageFunction and ShapedImageFunction flood-fill iterators in OnlyWritableIteratorsSet now use BinaryThresholdImageFunction; the spatial iterator keeps SphereSpatialFunction. MSVC instantiates the virtual IsPixelIncluded() with the class, which GCC and Clang do not.
• ITKCommon declares ITKImageFunction in TEST_DEPENDS for that header, with the usual "introduced by" comment; it was reachable only through ITKImageGrid's COMPILE_DEPENDS.
• The test's ITK includes are grouped ahead of the standard headers.

#7065's first five commits are these five again (git range-diff: all identical).

Local verification at 9229cb6

• ninja ITKCommonGTestDriver ITKDistanceMapGTestDriver (macOS arm64, Release): no compiler warnings.
• ctest -R 'FloodFillIterator|Reflective|ImageIterators': 22/22 passed.
• pre-commit run --all-files passes.
• The MSVC error cannot be reproduced on macOS; Windows CI is the check for it.