Add ADNI FDG preprocessing and release 4.4.0

[tctco]

Summary

• require --tracer {abeta,tau,fdg} for adni-pet-core
• preserve cerebellar-gray scaling for Aβ/tau and add ADNI's iterative masked global-mean normalization for FDG
• make tracer a required keyword in dcccpy.adni_pet_core()
• support the tracer branch for single, batch, BIDS, and automatic 4D motion-correction workflows
• document validation against four paired original/ADNI-preprocessed FDG studies and the small systematic difference in official DICOM scaling
• reduce pull-request CI to Linux x64 while retaining the full tag release matrix
• prepare DCCCcore 4.4.0, dcccpy 0.2.0, and runtime package 0.1.2 version metadata

Validation

• C++ Release build succeeds; DCCCcore --version reports 4.4.0
• 15 dcccpy tests pass
• affected ADNI CLI tests pass, including missing/invalid tracer, Aβ/tau, FDG fixed-point behavior, batch, BIDS, and 4D input
• four original six-frame FDG examples compared against date-matched ADNI set 3 results:
  • voxelwise correlation: 0.783–0.913
  • Dice at 0.5: 0.847–0.918
  • DCCC masked mean: 1.000000
  • ADNI example masked mean: 1.041–1.058

Release ordering

After merge, tag v4.4.0 and wait for the DCCCcore GitHub release assets to finish. Runtime packages must be built/published from those assets before publishing dcccpy 0.2.0.