Implement multiple parametrization for Negative Binomial

[theogf]

As discussed in #69, we need to be more flexible with the parametrization of NegativeBinomialLikelihood.
This is done here by having an extra parameter.
The interface is easily extended by defining a new type and two functions!

Also add docs to explain all this mess

There is still an issue with Functors.jl I will try to solve. But I thought I would already show the current status to get some feedback.

[codecov]

Codecov Report

Merging #80 (4c8b4d7) into master (498af31) will increase coverage by 0.74%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master      #80      +/-   ##
==========================================
+ Coverage   95.76%   96.50%   +0.74%     
==========================================
  Files          10       10              
  Lines         118      143      +25     
==========================================
+ Hits          113      138      +25     
  Misses          5        5              

Impacted Files	Coverage Δ
src/likelihoods/negativebinomial.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 498af31...4c8b4d7. Read the comment docs.

[st--]

Thanks; huge improvement!

Only thing you might want to add before merging is a comment on the parameter domains (i.e. successes and failures have to be positive, I think, \alpha definitely must be > 0 (will error for \alpha=0), don't know about \rho)?

[st--]

also, patch bump?