fix: set transform parameters by value when reconstructing from JSON

[vboussot]

CopyInFixedParameters/CopyInParameters std::copy into the transform's existing m_FixedParameters/m_Parameters without resizing them; the base class documents the requirement as "the range of values must conform to std::copy(begin, end, m_Parameters)". That holds for transforms that size those members in their constructor, but not for itk::BSplineTransform: its parameter count is unknown until the grid is set, and it keeps its coefficients in its own m_InternalParametersBuffer, so the base m_Parameters stays empty. Copying into it runs off the end of a zero-length buffer, which segfaults natively and aborts the module with no diagnostic under WebAssembly.

Use SetFixedParameters and SetParametersByValue instead, which every transform implements as a copy, and check the parameter count so a mismatch reports instead of corrupting memory. Fixed parameters are set first: for a B-spline they define the grid, and hence the number of parameters.

Fixed in both places that reconstruct a transform from TransformListJSON: itkWasmTransformToTransformFilter, in its composite and non-composite branches, and the copy of that pattern in the downsample package's resample pipelines.

Adds itkWasmTransformInterfaceBSplineTest, which round-trips a composite containing a B-spline through the interface. No existing test exercised a B-spline on this path, which is why the defect went unnoticed; the new test segfaults without this change.

Closes #1580

[thewtex]

🥇

[thewtex]

@vboussot thank you!

Since we changed the ITKWebAssemblyInterface module, we will need to first release a new docker base image, then release the downsample package built against this image. I will merge this after CI for the toolchain docker base images finishes green, then follow-up with those steps.