Add log-logistic distribution

[devmotion]

Updated and improved version of #1454.

Adding reference tests against R is a bit cumbersome currently, see #2007.
Edit: I added tests against R nevertheless by locally commenting out all other reference tests (and without environment, ie probably not reproducible).

Closes #1454.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 86.45%. Comparing base (0421b18) to head (4d24b07).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2008      +/-   ##
==========================================
+ Coverage   86.36%   86.45%   +0.08%     
==========================================
  Files         146      147       +1     
  Lines        8786     8842      +56     
==========================================
+ Hits         7588     7644      +56     
  Misses       1198     1198              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[devmotion]

I wonder if the order of parameters should be reversed to mimic e.g. Gamma and Weibull where the shape is the first and the scale parameter is the second argument. On the other hand, probably we don't have a completely consistent and clear convention currently - to solve such ambiguities it might be best to use keyword arguments (e.g., LogLogistic(; shape::Real, scale::Real)).

[andreasnoack]

Having shape and scale keywords would be wonderful. I look up the documentation every time I use Gamma. Then we could also have mean/dispersion, which in many cases would be even simpler to interpret. I think this was discussed in #823

[devmotion]

I tried something in #1405 but I didn't pursue it further since there was no response when I opened the PR. Probably, for most distributions exposing the somewhat arbitrary parameter names might also not be the best solution, maybe more meaningful names (such as shape and scale) might be better.