Supporting GLM fit with matrix for y

Currently [`dask_ml.linear_model.glm._GLM`'s `fit`]( https://github.com/dask/dask-ml/blob/v0.5.0/dask_ml/linear_model/glm.py#L141-L147 ) takes a matrix `X` and a vector `y`. However [`sklearn.linear_model.ElasticNet`'s `fit`]( https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/linear_model/coordinate_descent.py#L669-L678 ) (which is a base class for many other linear models in `scikit-learn`), takes a matrix `X` and a vector or matrix `y`. The latter is useful functionality for things like `SparseCoder` ( https://github.com/dask/dask-ml/issues/197 ), which use this to fit a sparse code to a given dictionary.