Combine repository specific labels with the universal model for issue…

… kind.

* To combine multiple models we define a base class IssueLabelModel
  that defines a common interface for all the models.

  * This way different models can just implement that interface and we
    can easily combine the results.

* UniveralKindLabelModel will be used for the generic model to predict
  label kind that is trained on all repositories.

  * The UniversalKindLabelModel class is based on the IssueLabeler code
    https://github.com/machine-learning-apps/Issue-Label-Bot/blob/536e8bf4928b03d522dd021c0464587747e90a87/flask_app/utils.py#L67

* kubeflow#70 Combine multiple models