Adds sklearn version check for ColumnTransformer import

[jrbourbeau]

This PR adds a scikit-learn version check to only add ColumnTransformer to the dask_ml.compose namespace if ColumnTransformer is present in sklearn (i.e. sklearn version >= 0.20.0.dev0).

For example, currently dask_ml will still attempt to import ColumnTransformer when using v0.19.1 of scikit-learn:

In [1]: import sklearn

In [2]: sklearn.__version__
Out[2]: '0.19.1'

In [3]: from dask_ml.compose import ColumnTransformer
---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
<ipython-input-3-037f94f5147b> in <module>()
----> 1 from dask_ml.compose import ColumnTransformer

~/github/dask/dask-ml/dask_ml/compose/__init__.py in <module>()
     16 # del parse
     17
---> 18 from ._column_transformer import ColumnTransformer, make_column_transformer
     19
     20 __all__ = ["ColumnTransformer", "make_column_transformer"]

~/github/dask/dask-ml/dask_ml/compose/_column_transformer.py in <module>()
      3 import numpy as np
      4 import pandas as pd
----> 5 import sklearn.compose
      6 from scipy import sparse
      7 from sklearn.compose._column_transformer import _get_transformer_list

ModuleNotFoundError: No module named 'sklearn.compose'

This sklearn version check is similar to what's done for OneHotEncoder in dask_ml/preprocessing/__init__.py

dask-ml/dask_ml/preprocessing/__init__.py

Lines 30 to 33 in b5b6954

	if SK_VERSION >= parse("0.20.0.dev0"):
	from ._encoders import OneHotEncoder # noqa
	__all__.append("OneHotEncoder")

[TomAugspurger]

Thanks!