BUG: Fix NRRD measurement-frame corruption; refactor reserved-field dispatch

[hjmjohnson]

Fixes a silent on-disk corruption in NrrdImageIO::Write and refactors the reserved-field routing. Documents the teem-driven NRRD version auto-bump that users on Discourse #6666 and in #5461 did not realise was already available.

Does not attempt to paper over user-supplied metadata in non-documented forms. Use NRRD_<field> per the (now-extended) class doxygen; non-conforming metadata either throws or falls through to NRRD's documented := custom-pair escape hatch.

Commit sequence

1. STYLE: convert NrrdImageIO::Write()'s metadata loop to a range-based for (mechanical keyIt-> → metaKey. rename, 27 / 27 lines).
2. STYLE: refactor the reserved-field if-ladder into a dispatch table keyed by teem's nrrdField enum. No behavior change.
3. ENH: document the teem-driven NRRD magic auto-bump (NRRD0001..NRRD0006) in the class doxygen and add an itkNrrdMetaDataTest subtest that writes a measurement frame via the documented NRRD_measurement frame key and asserts the on-disk magic is NRRD0005. Addresses the confusion behind Discourse #6666.
4. BUG: WriteMeasurementFrame now throws itk::ExceptionObject when the user-supplied frame's dimensions do not match nrrd->spaceDim, instead of padding missing entries with the magic sentinel 666666 (which teem faithfully wrote to the header, silently corrupting orientation data). Regression test added.

Relation to #5461 and SimpleITK/SimpleITK#2395

Neither is fixed here, by design:

• Write to NRRD doesn't convert all metadata keys to NRRD keys correctly #5461 — reporter used bare metadata keys like content and thicknesses instead of the documented NRRD_content / NRRD_thicknesses[N] form. The docs were corrected in DOC: Add information to NRRD and GDCM IOs. #5472; with those docs in place, the bare-key round-trip is user error. Non-NRRD_* keys go through NRRD's documented := custom-pair escape hatch, which is the standard's own answer for "anything not a reserved field." No code change in this PR is required for that contract to hold.

• setting NRRD metadata fails due to encoding information as string SimpleITK/SimpleITK#2395 — the Python binding stores numeric metadata as strings, which the ITK ExposeMetaData<double> then fails to coerce, leaving a zero. That is a Python-binding encoding bug in SimpleITK, not a NRRD writer bug. Fixing it inside the C++ writer would require type-coercion machinery that accommodates a specific wrapper's encoding choice — maintenance burden for the core codebase in service of a third-party workaround. The upstream SimpleITK issue is the correct venue.

Test plan

• ctest -R Nrrd — 33/33 pass locally (Linux, clang 19 via conda-forge).
• itkNrrdMetaDataTest extended with two new subtests: NRRD0005 auto-bump from NRRD_measurement frame; throw-on-malformed-measurement-frame regression.
• Verify on Windows/macOS runners once CI picks up the branch.

[greptile-apps]

Greptile Summary

This PR fixes a silent on-disk corruption bug in NrrdImageIO::Write where a user-supplied measurement frame with wrong dimensions was padded with the sentinel value 666666 and written verbatim to the NRRD header, corrupting orientation data. The fix throws itk::ExceptionObject on dimension mismatch and refactors the old strncmp/sscanf if-ladder into a clean dispatch table keyed by teem's nrrdField enum. Doxygen and a new regression test (malformed frame + NRRD0005 auto-bump) are also added.

Confidence Score: 5/5

Safe to merge; the core bug fix and dispatch refactor are correct, and the only remaining finding is a minor Windows test-cleanup issue.

All prior P0/P1 concerns (666666 corruption, warning macro routing, TryExposeDouble return value, readNrrdHeader binary mode) were addressed in earlier commits per the developer replies. The sole new finding is a P2 test-cleanup portability issue that does not affect correctness.

itkNrrdMetaDataTest.cxx — open ifstream during RemoveFile call may leave a stale temp file on Windows.

Important Files Changed

Filename	Overview
Modules/IO/NRRD/src/itkNrrdImageIO.cxx	Core bug fix: replaces the old 666666-sentinel corruption with a proper dimension check + exception in WriteMeasurementFrame; refactors the per-field if-ladder into a clean dispatch table — logic and validation are correct.
Modules/IO/NRRD/include/itkNrrdImageIO.h	Documentation-only: adds a doxygen paragraph explaining teem's automatic NRRD version selection (NRRD0001–NRRD0006). No behavioral changes.
Modules/IO/NRRD/test/itkNrrdMetaDataTest.cxx	Two new regression subtests added (NRRD0005 auto-bump and throw-on-malformed-frame); a minor Windows portability issue with an open ifstream during RemoveFile.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[NrrdImageIO::Write - MetaDataDictionary loop] --> B{Key starts with 'NRRD_' prefix?}
    B -- Yes --> C[Extract keyField after prefix]
    C --> D[TryDispatchNrrdReservedField]
    D --> E{Match in kNrrdReservedFields dispatch table?}
    E -- Per-axis field --> F[Parse index via sscanf, call perAxis handler, stamp nrrd->axis]
    E -- Scalar field --> G{Field == measurement frame?}
    G -- Yes --> H{Frame dims == nrrd->spaceDim?}
    H -- No --> I[itkGenericExceptionMacro throw ExceptionObject]
    H -- Yes --> J[Copy frame to nrrd->measurementFrame]
    G -- No --> K[WriteOldMin / WriteOldMax / WriteSpace / WriteContent / etc.]
    E -- No match --> L[Key silently consumed, not forwarded to custom-pair]
    B -- No --> M[Serialize value to string, nrrdKeyValueAdd as := pair]

Loading

_{Reviews (2): Last reviewed commit: "BUG: Throw on malformed NRRD measurement..." | Re-trigger Greptile}

[hjmjohnson]

@greptileai review this draft before I make it official