Support thermal noise removal with dwidenoise and/or NORDIC

[tsalo]

The main questions are:

1. Will fMRIPrep incorporate NORDIC? If so, then there's no need to implement it in fMRIPost-Phase.
2. Does NORDIC meaningfully impact preprocessing (i.e., motion parameters, boldref space)? If so, then we really need it in fMRIPrep. Otherwise, we can apply it to the raw data in fMRIPost-Phase, then run STC, and then apply the transforms from fMRIPrep.

Related issues/PRs: SteenMoeller/NORDIC_Raw#21, nipreps/fmriprep#3308.

[tsalo]

Even if we add NORDIC to fMRIPrep, it won't output processed phase data, so we need to run NORDIC with the same parameters in fMRIPost-Phase to get the denoised phase data, right? Assuming NORDIC actually modifies the phase data of course.

[tsalo]

@mattcieslak had the great idea to test out the effect of NORDIC on motion parameters by simulating data with POSSUM, although he already found that MP-PCA had a large effect on dMRI motion parameters and I'm betting that extends to fMRI.

[tsalo]

I think I'll just use my Python translation internally within fMRIPost-Phase. At least until dwidenoise has it implemented without any licensing issues.

[tsalo]

It should be good to go now.