WIP (Not ready): Add GradientImageFilter::OverrideBoundaryCondition(std::unique_ptr<BoundaryConditionType>)

[N-Dekker]

Deprecated the old GradientImageFilter member function OverrideBoundaryCondition(BoundaryConditionType *) in favor of OverrideBoundaryCondition(std::unique_ptr<BoundaryConditionType>).

Aims to make it more clear that the filter takes the ownership of the specified BoundaryCondition object.

Triggered by a comment from Bradley Lowekamp (@blowekamp) at #4512 (comment)

[N-Dekker]

For the record, I started a discussion on how unique_ptr parameters are supported by SWIG, at:

• Does SWIG support passing an `std::unique_ptr` parameter "by-value"? swig/swig#2854

[hjmjohnson]

@N-Dekker — I've opened #6714, which carries this idea forward. Thank you for it, and apologies it sat blocked this long: the diagnosis here was correct, and the design you proposed is the one that landed.

The SWIG blocker is gone. ITK now pins SWIG 4.3.0, which ships std_unique_ptr.i. I verified the whole chain rather than assuming it: castxml emits the method, pygccxml resolves the argument as ::std::unique_ptr<...> (not ?unknown?, so igenerator.py's skip_method no longer drops it), the %unique_ptr typemap binds, and the generated C++ compiles, links, and passes a runtime ownership test — the exact step that failed in 2024.

One trap worth recording for anyone doing this elsewhere: %unique_ptr must be given ITK's mangled alias (itkImageBoundaryConditionIF2), not the C++ spelling. The C++ spelling fails silently — SWIG falls back to the default typemap with no error or warning.

What #6714 changes relative to this PR

• Same core idea, but the new method is named SetBoundaryCondition and the existing OverrideBoundaryCondition(T *) is deprecated under ITK_FUTURE_LEGACY_REMOVE rather than changed in place, so existing callers keep both compiling and their current ownership semantics.
• Widened to the whole family. GradientImageFilter was the only one of six classes whose identically-signed OverrideBoundaryCondition took ownership; the other five now say explicitly in their docs that they do not. The rule: OverrideBoundaryCondition(T *) never takes ownership, and any owning API uses a different name and a unique_ptr parameter.
• Adds GetBoundaryCondition / ResetBoundaryCondition, GTests that observe ownership via a destruction counter, a Python test for the wrapped behavior, and a migration-guide entry.

I'd suggest closing this PR in favor of #6714, but that's your call — I haven't automated it, and there's no Closes keyword in the new commit.