Implement Nuisance Regression

[nx10]

Remove confound signals with 36-parameter and aCompCor options.

Regressors (src/rbc/core/functional/regressors.py):

• Motion: 6 params + derivatives + squared + deriv² = 24
• CSF: mean + derivatives + squared + deriv² = 4
• WM: mean + derivatives + squared + deriv² = 4
• Global: mean + derivatives + squared + deriv² = 4 (36-param)
• aCompCor: 5 PCs from WM+CSF (aCompCor set)

Mask erosion (src/rbc/core/functional/mask_utils.py):

• CSF: 90% volume, WM: 60% volume, Brain: 30mm

Regression:

• AFNI 3dTproject with bandpass 0.01-0.1 Hz
• Unit tests (derivatives, regressors, erosion)
• Integration test (~2-3 min)

Outputs: desc-cleaned_bold (36-param and aCompCor versions)

Requires: Issue #3 (CSF/WM masks), Issue #5 (motion params)

Reference: Step 15: Nuisance Regression