[FIX] Use linalg.eigh in RESS

[nbara]

This replaces linalg.eig with linalg.eigh in the RESS code (as it provides slightly better results and the symmetry assumption for eigh is valid with covariance matrices)

Had to use a tiny bit of shrinkage to the covariance to avoid numerical precision issues in the GED.

Thanks @mikexcohen for the suggestion.

Linking this cool new preprint here as well:
https://arxiv.org/abs/2104.12356

[codecov]

Codecov Report

Merging #46 (c154600) into master (f2fe945) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #46      +/-   ##
==========================================
+ Coverage   78.22%   78.24%   +0.01%     
==========================================
  Files          20       20              
  Lines        2209     2211       +2     
==========================================
+ Hits         1728     1730       +2     
  Misses        481      481              

Impacted Files	Coverage Δ
meegkit/asr.py	85.02% <100.00%> (ø)
meegkit/dss.py	81.94% <100.00%> (ø)
meegkit/ress.py	100.00% <100.00%> (ø)

---

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 f2fe945...c154600. Read the comment docs.