Optimizing Perrone example

[Neeratyoy]

Reference Issue

Addresses #852.

What does this PR implement/fix? Explain your changes.

Removes the repeated calls to openml.runs.get_runs and replaces it with one single API call using openml.evaluations.list_evaluations_setups.
Also, added a unit test for this example.

[codecov-io]

Codecov Report

Merging #853 into develop will decrease coverage by 0.09%.
The diff coverage is n/a.

@@            Coverage Diff            @@
##           develop    #853     +/-   ##
=========================================
- Coverage    88.49%   88.4%   -0.1%     
=========================================
  Files           37      37             
  Lines         4269    4269             
=========================================
- Hits          3778    3774      -4     
- Misses         491     495      +4

Impacted Files	Coverage Δ
openml/datasets/functions.py	94.06% <0%> (-1.25%)	⬇️

---

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 f74b73a...dc1ce33. Read the comment docs.

[mfeurer]

Thanks a lot, this looks really good. However, could you please explain why you need an extra test given that the example is executed by travis-ci?

[Neeratyoy]

However, could you please explain why you need an extra test given that the example is executed by travis-ci?

I came across the issue #805 and thought this was a good opportunity to incorporate it. Also, I came across a unit test for the SVM example. Hence, I thought this was the protocol.
Having the example as a unit test also seemed to be a good check since from what I understand, we won't know if the example script failed, in terms of an error from travis-ci, will we? That is, we might still get all greens.