ENH: Allow ingestion of alternatively grouped CompositeH5s

[mgxd]

ITKCompositeH5 is quite rigid ATM, bypassing the first H5 group entirely. This can be fine in some cases, but I have a transform containing a single displacement field (generated from a composite transform and reference image) which is not being read.

This PR successfully loads the DenseFieldTransform. If this looks good, can work on tests.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.33%. Comparing base (4d7ae6c) to head (08c3c84).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #233   +/-   ##
=======================================
  Coverage   95.32%   95.33%           
=======================================
  Files          28       28           
  Lines        2954     2960    +6     
=======================================
+ Hits         2816     2822    +6     
  Misses        138      138           

Flag	Coverage Δ
unittests	95.11% <100.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[effigies]

How did you generate the h5? With ants, the first group was just metadata.

Probably good to be flexible anyway, but if we're generating with Python, we might need to be careful to emulate itk.

[mgxd]

It is with ants (ApplyTransforms)

https://github.com/nipreps/nibabies/blob/86264fee0b56144d0799b29a445f84e20ea733f9/nibabies/interfaces/patches.py#L72-L73

$ antsApplyTransforms --help
     -o, --output warpedOutputFileName
                  [warpedOutputFileName or compositeDisplacementField,<printOutCompositeWarpFile=0>]
                  Linear[genericAffineTransformFile,<calculateInverse=0>]
          One can either output the warped image or, if the boolean is set, one can print 
          out the displacement field based on the composite transform and the reference 
          image. A third option is to compose all affine transforms and (if boolean is 
          set) calculate its inverse which is then written to an ITK file.