Develop a toolkit to transform features according to a feature configuration.

[workingloong]

It is boring for users to develop feature transformation using Keras layers or feature columns when the features count is very large. We can provide a configuration class for uses to define their transformation logic like:

Class FeatureTransformConfig:
    standardized_features = []
    bucketized_features = []
    round_features = []
    hash_features = []
    lookup_features = []
    loground_features = []
    feature_groups = []

We should take the transformation DAG in #1856 to design the configuration format.

The toolkit can transform inputs according to the configuration and output the transformation result.

def transfrom(inputs, transform_config):
"""
Args:
    inputs: a dict with `tf.keras.layer.Input`
    transform_config: feature transform configuration

Returns:
   A dict: the name is feature name and the value is the transformation result.
"""