DateTimeFeaturizer should return integer-valued features

[freddyaboulton]

The DateTimeFeaturizer returns string-valued features for month and day of week. These are useful features but the string values necessitate having an encoder downstream in the pipeline. The DateTimeFeaturizer should handle the encoding and provide a mapping so users know how the encoding happened.

Proposed output:

New api method:

dt = DateTimeFeaturizer()
dt.fit(X, y)
dt.get_feature_names()

{"Date_month: ["January", ..., "December"],
 "Date_day_of_week: ["Monday, ..., "Sunday"]
}