Fix for one tiny aspect of the rank deficiency issues

[palday]

This fixes one tiny facet of the bigger issue lurking in #324.

In particular, OpenBLAS on SkyLakeX can do something weird for some (full rank!) matrices:

• The pivoted Cholesky estimates rank below full.
• The unpivoted Cholesky has info ==0, which means that the unpivoted Cholesky succeeded, suggesting numerical full rank.

In this case, the old code generated an invalid permutation consisting of the identity permutation with an extra element 0 appended to the end. The fix here trusts the success of the unpivoted Cholesky more than the rank of the pivoted Cholesky.

[dmbates]

Looks good. Let's wait for the checks to run then feel free to merge.

[codecov]

Codecov Report

Merging #367 into master will not change coverage.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master     #367   +/-   ##
=======================================
  Coverage   95.40%   95.40%           
=======================================
  Files          23       23           
  Lines        1503     1503           
=======================================
  Hits         1434     1434           
  Misses         69       69           

Impacted Files	Coverage Δ
src/linalg/statschol.jl	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 9b7a748...d3f85e2. Read the comment docs.

[dmbates]

Is the failure on 1.5 a problem? The failure on nightly is because Showoff hasn't been updated for the Grisu -> Ryu that took place back in the Julia-1.4.0 release.

[palday]

The failure for 1.5 is the main rank deficiency issue (#324) and the failure for nightly is the Showoff issue (#363).

[palday]

It's not much comfort, but on my other Linux machine without SkyLakeX, the tests pass on 1.5.

(I have learned far too much about numerical linear algebra and its interaction with different SIMD instructions for this issue ... and yet I still know that I know nothing.)

[dmbates]

It's not much comfort, but on my other Linux machine without SkyLakeX, the tests pass on 1.5.

(I have learned far too much about numerical linear algebra and its interaction with different SIMD instructions for this issue ... and yet I still know that I know nothing.)

One of Piet Hein's "grooks" is

Knowing what thou knowest not is, in a sense, omniscience.