fix stacking

[yuta100101]

fixed some bugs of stacking().

1. pass type_of_target argument of stacking() to cross_validate().
   • By default, the type_of_target of cross_validate() is determined by sklearn.utils.multiclass.type_of_target. Because of this, even if you set type_of_target="continuous" for regression with integer targets, cross_validate()'s type_of_target is "multiclass".
2. fixed default estimator for regression.
   • to adapt to removal of normalize argument
   • related issue
   • add unittest.
3. fixed default estimator for classification.
   • add solver="liblinear" to adapt to change of the default value.
   • related article