ENH: implement the hierarchical clustering analysis of within-scanner SC reliability

[celprov]

I have implemented the analyses we promised under the section "hierarchical clustering analysis of within-scanner SC reliability" of our registered report.
For now, the code runs on simulated SC matrices or on the FC matrices loaded from HCPh.
Specifically, I implemented:

1. aggregating the SC matrices in an array amenable for clustering analysis
2. perform hierarchical clustering using Scikit-learn as pre-registered and plot the dendrogram
3. extract average head motion, phase encoding direction, time of day, and room temperature to investigate whether such confounds explain similarity between SC matrices
4. perform tSNE; the documentation of tSNE mention it is recommended to reduce dimensionality with PCA to approx 50 features if nbr of feature is too high (2016 in the case of the lowest DiFuMo scale), so we need implement this.
5. Load the confounds and color the dendogram and the tSNE embedding according to factor values.