Fix bugs in robust module and improve test coverage

[TimotheeMathieu]

In PR #76 I succeeded in making examples faster but as a consequence some of the examples do not work anymore.
It allowed me to see some bugs in the robust module and this PR is to resolve them and try to improve the test in order not to encounter this situation again.

TODO:

• add test for the weights_ parameter
• add test to compare to already existing robust algo in regression in sklearn (TheilSen, Ransac)
• add tests
• fix examples

If anyone has a test suggestion, please share in comment.

[TimotheeMathieu]

@rth the PR is ready for review, thanks in advance. For now I removed the real data classification example as I didn't find a good example to replace the last one that had some problems.

[rth]

Thanks @TimotheeMathieu ! Generally looks good to me aside for the two following comments.

[TimotheeMathieu]

I fixed the test for kmedoids here mentioned in PR #73

EDIT : I removed the test in order to make a separate PR.

[TimotheeMathieu]

@rth I don't understand the error in azure pipeline for python 3.9.
I can't reproduce it in a virtual env so I guess it is a problem with the azure pipeline.

[adrinjalali]

The azure-pipelines.yaml needs fixing. The latest scipy release supports python3.9 already, shouldn't be fetching the nightly release. Not sure what's making it fetch from the nightly repo.

The scikit-learn pre-release also supports py3.9 already.

[rth]

Sorry I won't be able to do a detailed review, but generally looks good to me. Thanks!

Feel free to merge once CI is green.