COMP: Build against ITK 6

[hjmjohnson]

Build against ITK 6 while keeping ITK 5.4: select the PocketFFT 1D filters where ITK 6 removed the VNL FFT backend, and call pocketfft directly in Spectra1DImageFilter, which used vnl_fft_1d on a vnl_vector.

Details

• ITK 6 removed the VNL FFT backend; the Vnl*1DFFTImageFilter names survive only as deprecation shims that a legacy-free build compiles out, and vnl/algo/vnl_fft_1d.h is gone. These classes reach FFTImageFilterFactory as template template arguments keyed on the template itself, so the selection is by macro rather than by alias template.
• Spectra1DImageFilter called vnl_fft_1d::bwd_transform, for which no ITK filter substitutes. The pocketfft call follows ITK's own removed vnl_fft_1d shim (a4de8b36f76): vnl dir=+1 is exp(+i), pocketfft's backward direction, so bwd_transform is a forward pocketfft transform with no normalization.

Verification

Configured with ITK_DIR pointing at an installed ITK 6 tree (upstream main as of 2026-09-22, ITK_WRAP_PYTHON=ON, with the ingested modules this module depends on enabled), built the module and its Python wrapping. The module still builds against ITK 5.4.

[dzenanz]

Changes look good. But my nightly builds run into VTK-related configure error.

[hjmjohnson]

Pushed 0501b9d: the tests exercised the Vnl 1D FFT filters by name as an explicit backend (backend == 1); on ITK 6 that backend is PocketFFT, selected with the same version guard as src/. The module, its examples, tests, and header test now build against an installed ITK 6, and 37 of 48 tests pass. The 11 that fail are not caused by this branch:

• itkSpectra1DImageFilterTest: the reference spectrum input rf_voltage_15_freq_0005000000_2017-5-31_12-36-44_ReferenceSpectrum_side_lines_03_fft1d_size_128.mha resolves (CID bafkreigfnnwh…) to a 479-byte file that is only a MetaImage header, so the reader fails with a zlib error. The published data object appears truncated.
• The ten Python* tests: itk_python_add_test puts ${ITK_DIR}/Wrapping/Generators/Python first on PYTHONPATH, which against an installed ITK is the wrapping infrastructure rather than ITK's Python package. Tracked as a follow-up on ENH: Build, test, and wrap external modules against an installed ITK InsightSoftwareConsortium/ITK#6891; these tests pass in a build-tree configuration.

[dzenanz]

I opened follow-up PR #253 with that commit.