Use 3D fftconvolve instead of slice-wise 2D fftconvolve

[notZaki]

This PR replaces the multiple slice-wise 2D fftconvolve with a single 3D fftconvolve to make the code simpler. The original implementation and the PR produce a nearly identical array.

• Major difference
  • The first and last slice is zero in the original implementation, i.e. data_corr[:,:,0] and data_corr[:,:,-1] are all zeros
    • This doesn't seem realistic, and could potentially interfere with the fft step that happens near the end of ent_rate_sp()
  • The PR version has non-zero values in the first and last slice
• Minor difference
  • The original implementation is symmetric around the midpoint of the z direction, i.e. data_corr[:,:,50] == data_corr[:,:,-51]
  • The PR version is symmetric, but flipped, i.e. data_corr[:,:,50] ≈ np.flip(data_corr[:,:,-51])

[codecov-io]

Codecov Report

Merging #15 (87527e8) into main (67ad715) will decrease coverage by 0.16%. The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main      #15      +/-   ##
==========================================
- Coverage   92.09%   91.92%   -0.17%     
==========================================
  Files           3        3              
  Lines         291      285       -6     
==========================================
- Hits          268      262       -6     
  Misses         23       23              

Impacted Files	Coverage Δ
mapca/utils.py	98.54% <100.00%> (-0.07%)	:arrow_down:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 67ad715...87527e8. Read the comment docs.

[eurunuela]

I would like to add an integration test before we merge PRs like this one. I was thinking of saving the results before and after maPCA in tedana and using those.

[eurunuela]

@tsalo feel free to merge this PR if you're happy with the changes.