Reduce memory usage of CBMAEstimator's FWE correction

[tsalo]

Closes None. A user reported memory errors when running an ALE with voxel-level FWE correction on a ~2k-study subset of the NeuroQuery database. I've been trying to figure out what the problem is and how to fix it, and this is the best I've come up with so far.

EDIT: It still doesn't address the problem.

Changes proposed in this pull request:

• Determine the random coordinates for each permutation within the permutation function, rather than defining all of them at once and then splitting them up for the loop.
  • I am pretty sure that this will slow things down, since the previous approach was something we optimized for speed. Unfortunately, it just doesn't scale well for very large Datasets.

[tsalo]

I have no clue why CI isn't triggering...

[jdkent]

could be helpful for profiling: https://github.com/bloomberg/memray

[codecov]

Codecov Report

Merging #659 (99f7616) into main (6cd751e) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main     #659   +/-   ##
=======================================
  Coverage   85.28%   85.29%           
=======================================
  Files          40       40           
  Lines        4513     4514    +1     
=======================================
+ Hits         3849     3850    +1     
  Misses        664      664           

Impacted Files	Coverage Δ
nimare/meta/cbma/base.py	96.53% <100.00%> (+0.01%)	⬆️

---

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 b818512...99f7616. Read the comment docs.

[adelavega]

@tsalo given #676 is this stale / closable?

[tsalo]

Yeah, I think it can be closed.