fix example: sklearn cross_validation module EOL

[shiodat]

The sklearn.cross_validation module is EOL in sklearn 0.20, which is now stable.
The classifier_kfold.py example uses it and I fixed it to use sklearn.model_selection module.

[coveralls]

Coverage remained the same at 88.032% when pulling 32e432b on fix-sklearn-eol-functions into 70d920b on develop.

[rimms]

👍 LGTM

• scikit-learn 0.17.1

  $ pip freeze | grep scikit-learn
  scikit-learn==0.17.1
  $ coverage run -p tools/run_examples.py
  
    .. snip ..
  
  --- classifier_kfold.py -----------------------------
  
    .. snip ..
  
    virginica       1.00      0.02      0.04        50
  
  avg / total       0.84      0.67      0.57       150
  

• scikit-learn 0.20.0

  $ pip freeze | grep scikit-learn
  scikit-learn==0.20.0
  $ coverage run -p tools/run_examples.py
    .. snip ..
  
  --- classifier_kfold.py -----------------------------
  
    .. snip ..
  
     virginica       1.00      0.24      0.39        50
  
     micro avg       0.75      0.75      0.75       150
     macro avg       0.86      0.75      0.70       150
  weighted avg       0.86      0.75      0.70       150