Add primitives to use Ballet together with MLBlocks

[micahjsmith]

• the ballet.engineer_features primitive applies a given project's feature engineering pipeline
• the ballet.encode_target primitive applies a given project's target encoder
• two sample pipelines are included that put these two steps together with either RF classifier or regressor
• a ballet.decode_target primitive may be necessary in the future but is not implemented here: (1) the target encoder is not necessarily reversible (2) the primitive needs to be able to restore its state from the context similar to class decoder

Example usage (for predict-census-income):

import mlblocks
from ballet import b
from sklearn.metrics import classification_report

X_df, y_df = b.api.load_data()
X_df_te, y_df_te = b.api.load_data(input_dir='data/val')

encoder = b.api.encoder
y = encoder.fit_transform(y_df)
y_te = encoder.transform(y_df_te)

pipeline_info = mlblocks.load_pipeline('ballet_rf_classifier')
pipeline = mlblocks.MLPipeline(pipeline_info)

pipeline.fit(X_df, y_df)

y_pred = pipeline.predict(X_df)
report = classification_report(y, y_pred, output_dict=True)

y_pred_te = pipeline.predict(X_df_te)
report_te = classification_report(y_te, y_pred_te, output_dict=True)