ENH: Ingest AnisotropicDiffusionLBR into Modules/Filtering

[hjmjohnson]

Moves AnisotropicDiffusionLBR from a configure-time remote fetch into the ITK source tree at Modules/Filtering/AnisotropicDiffusionLBR/ using the v3 whitelist ingestion strategy. Scope is deliberately narrow: only code, headers, tests, wrapping, and module build descriptors cross the merge boundary. Everything else (Old/, examples/, paper/, docs, packaging/CI scaffolding) stays in the archived upstream repo. Relates to #6060.

Strategy: v3 whitelist + CID normalization

Evolved across two feedback rounds in this PR:

• @blowekamp (2026-04-21): full-history merges risk git-object bloat; Old/, paper material, demo PNGs should not enter ITK.
• @dzenanz (2026-04-21): full history where clean, squash where messy; migrate all hash-based content-links to .cid; examples belong in top-level Examples/, not inline with module source.

The v1 approach (full-history merge of everything upstream shipped) was abandoned. The v2 bloat-thresholded approach was superseded by the v3 structural whitelist: a git filter-repo --paths pass that admits only the whitelisted categories. Strategy documents live at .claude/worktrees/pr-6061-local-beta-modules-support/ (INGESTION_STRATEGY.md, AUDIT_DESIGN.md, CLEANUP_CHECKLIST.md).

What ingested (whitelist)

The git filter-repo pipeline that produced the merge commit:

git filter-repo
  --path include --path src --path test --path wrapping
  --path CMakeLists.txt --path itk-module.cmake
  --to-subdirectory-filter Modules/Filtering/AnisotropicDiffusionLBR
  --prune-empty always
git filter-repo --invert-paths --path '.../CTestConfig.cmake'  # second pass

Result: 168 upstream commits → 106 surviving on the filtered branch; 43 files transferred; 15 distinct authors preserved; git blame walks across the merge boundary into upstream authors.

What did NOT ingest (see Modules/Filtering/AnisotropicDiffusionLBR/README.md)

• Old/ — pre-refactor legacy trees
• examples/ + examples/Data/ — per @dzenanz, per-module examples are routed to top-level Examples/ in a separate follow-up PR, not ingested inline
• README.rst — useful content is folded into Doxygen on the filter headers
• CTestConfig.cmake, pyproject.toml, .github/, .clang-format, LICENSE — standalone-repo scaffolding superseded by ITK root equivalents

The new in-tree Modules/Filtering/AnisotropicDiffusionLBR/README.md documents each exclusion with a pointer at the archived upstream URL.

Commit structure (4 commits)

1. ENH: Ingest ITKAnisotropicDiffusionLBR into Modules/Filtering — merge commit of the whitelist-filtered upstream
2. DOC: Add README.md pointing at archived upstream for AnisotropicDiffusionLBR — pointer doc + exclusion rationale
3. COMP: Remove AnisotropicDiffusionLBR.remote.cmake; now in-tree — delete the remote fetch declaration
4. ENH: Enable AnisotropicDiffusionLBR in CI via configure-ci — opt the module into the Pixi CI matrices

⚠ Pre-merge step: CID normalization pending

Per the v3 strategy, the 24 test-data content-links under test/Baseline/ and test/Input/ should be normalized from .md5 to .cid before this PR merges. The conversion is mechanical — resolve each MD5 hash to its stored bytes via the ExternalData mirror, compute the CIDv1, write <file>.cid, delete <file>.md5.

This step was not executed in the ingest session because the session lacked network access to data.kitware.com and web3.storage. Either:

• A maintainer with network access runs Utilities/SPDX/ (no — wrong tool), or more accurately a new Utilities/Maintenance/cid-normalize.sh helper (follow-up) that invokes @web3-storage/w3cli per Documentation/docs/contributing/upload_binary_data.md;
• OR we accept the .md5 content-links as-is for this PR and open a follow-up sweep PR that converts .md5 → .cid across the whole tree (not just this module), matching ITK's own precedent at commit f3899ce8c6.

The tree-wide sweep is probably the less disruptive path — reviewers of this ingest PR can focus on the structural ingest without a CID-conversion diff on top.

AI disclosure

Ingestion pipeline (filter-repo pass order, README scope, commit structure) developed with Claude Code assistance. Every commit locally rebuildable and testable; git log shows the 15 upstream authors preserved across the whitelist-filtered merge.

[hjmjohnson]

@greptileai review this draft before I make it official.

The previous review was against an older tip; the branch has since been force-pushed through multiple rounds (v3 pipeline re-ingest, gersemi STYLE, KWStyle private: fix in itkLinearAnisotropicDiffusionLBRImageFilter.hxx). Current tip: 5a46182206. All CI platforms now green except the accepted ghostflow-check-main.