Calculate dependence metrics for modified mixing matrices and confounds

[tsalo]

Closes #1300 and closes #1341.

Changes proposed in this pull request:

• Run generate_metrics on the following:
  • The orthogonalized mixing matrix if --tedort is enabled.
  • The post-MIR mixing matrix if --gscontrol mir is enabled.
  • Any external regressors provided by the user.
  • The gsr and mir time series if --gscontrol gsr and/or mir are enabled.
• Add use_multivariate argument to generate_metrics. It should be true for mixing matrices and false for external regressors and confounds. It shouldn't matter for PCA, since orthogonal components should produce the same component weights whether or not a multivariate model is used.

[codecov]

Codecov Report

❌ Patch coverage is 92.30769% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.36%. Comparing base (85fab4f) to head (a8b4113).
⚠️ Report is 3 commits behind head on main.

Files with missing lines	Patch %	Lines
tedana/metrics/dependence.py	81.48%	4 Missing and 1 partial ⚠️
tedana/metrics/collect.py	50.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1350      +/-   ##
==========================================
+ Coverage   90.28%   90.36%   +0.08%     
==========================================
  Files          30       30              
  Lines        4712     4776      +64     
  Branches      784      791       +7     
==========================================
+ Hits         4254     4316      +62     
- Misses        309      312       +3     
+ Partials      149      148       -1     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[tsalo]

I can work on the test coverage once we decide if we want to move forward with this.

[tsalo]

One thing I learned from trying out this branch on real data is that the dependence metrics will change only for accepted components when using the orthogonalized mixing matrix. Given how orthogonalization is meant to be used (i.e., you would do denoising using only the orthogonalized rejected components), it might make more sense to calculate the dependence metrics on a mixing matrix that only contains the orthogonalized rejected components.

[tsalo]

I tried generating the metrics using just the orthogonalized rejected components, and the dependence metrics stayed the same as the original values for those components! I'm a little surprised.

[tsalo]

Would figures like this be useful to anyone?