Revert "Add numerically safe ordered probit distribution."

[Spaak]

Reverts #4232 . Although all tests were fine within this PR branch, we now have a failure on master, see https://github.com/pymc-devs/pymc3/runs/1562295314 .

Any input on how to avoid such issues would be very helpful :) @MarcoGorelli perhaps? For now I think reverting the merge is probably the best option? (Which is what this PR does.)

Ping @tohtsky as well, who wrote #4232 . (Not to assign blame! You could not have foreseen this, given the OK test status on the PR.)

[michaelosthege]

@Spaak I don't think we should revert. I had the same flaky test on #4214 but it went away after re-running the CI jobs.
So first just hit "re-run all jobs" on the master CI and it might turn green already.

[codecov]

Codecov Report

Merging #4345 (e47c4c0) into master (9dbf9ea) will increase coverage by 0.01%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #4345      +/-   ##
==========================================
+ Coverage   87.85%   87.87%   +0.01%     
==========================================
  Files          88       88              
  Lines       14501    14473      -28     
==========================================
- Hits        12740    12718      -22     
+ Misses       1761     1755       -6     

Impacted Files	Coverage Δ
pymc3/distributions/__init__.py	100.00% <ø> (ø)
pymc3/distributions/discrete.py	96.78% <ø> (+1.56%)	⬆️
pymc3/distributions/dist_math.py	91.85% <ø> (-0.12%)	⬇️

[MarcoGorelli]

So first just hit "re-run all jobs" on the master CI and it might turn green already.

Let's just make sure to leave some kind of comment/issue if we do this, flaky tests should be fixed eventually

[Spaak]

Re-running CI on master now, let's hope the test failure indeed proves flaky and goes away. I'll leave this PR open until we know.

[Spaak]

Yep, indeed CI on master is now all green. See #4323 for the issue about flaky tests.