make DFTransformer inherit BaseEstimator

[Qi-max]

Summary

Make DFTransformer inherit BaseEstimator

• get_params and set_params can be used.
• print(MatPipe()), as an example, can give somewhat better output (good for demo):

previously:

print(MatPipe())
<automatminer.pipeline.MatPipe object at 0x12a14cfd0> 

now:

print(MatPipe())
MatPipe(autofeaturizer=AutoFeaturizer(bandstructure_col=None, cache_src=None,
        composition_col='composition', do_precheck=True, dos_col='dos',
        drop_inputs=True, exclude=[],
        featurizers={'composition': [AtomicOrbitals(), ElementProperty(data_source=<matminer.utils.data.PymatgenData obj...er (INFO)>,
        multiindex=False, n_jobs=None, preset='best',
        structure_col='structure'),
      cleaner=DataCleaner(drop_na_targets=True, encode_categories=True, encoder='one-hot',
        feature_na_method='drop', logger=<Logger automatminer (INFO)>,
        max_na_frac=0.01, na_method_fit='drop', na_method_transform='fill'),
      learner=TPOTAdaptor(logger=<Logger automatminer (INFO)>),
      log_level=None, logger=<Logger automatminer (INFO)>,
      reducer=FeatureReducer(corr_threshold=0.95, keep_features=None,
        logger=<Logger automatminer (INFO)>, n_pca_features='auto',
        n_rebate_features=0.3, reducers=('pca',), remove_features=None,
        tree_importance_percentile=0.9))

[ardunn]

Thanks @Qi-max seems like a good idea to me, though I will need to run full tests (commit CircleCI tests are not the full ones) with this before merging since it changes every single part of MatPipe (despite being only one line, I am not sure what could go wrong with such a fundamental change).

Also would you mind adding a quick few tests to make sure the BaseEstimator behavior for get_params and set_params is correct? These should run super fast cuz all you have to do is make a matpipe and get/swap out params and test against attributes to make sure the parameters are correct, you don't actually have to do any ML.