rf: Reconstruct motion confounds from minimal derivatives

[effigies]

After an exploration in nipreps/fmripost-aroma#94, we now know how to reconstruct the original motion parameters (to at least 0.001 mm/rad precision) from the ITK transforms we generate.

To ensure consistency between single-shot and fit+apply runs, I propose to use this function in place of the original outputs.

I will add a regression test before merging, to ensure we notice if a change in nitransforms breaks this calculation.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 72.70%. Comparing base (c3cbe55) to head (da5a025).
Report is 10 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3424      +/-   ##
==========================================
+ Coverage   72.33%   72.70%   +0.37%     
==========================================
  Files          57       57              
  Lines        4269     4327      +58     
  Branches      545      546       +1     
==========================================
+ Hits         3088     3146      +58     
  Misses       1065     1065              
  Partials      116      116              

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

[effigies]

Timeseries are being generated, the outputs are close to those from master, and tests confirm the validity of the changes.