Improve inference of numeric categorical values

By default Woodwork will not infer columns with a numeric dtype to be Categorical. These columns will only be inferred as Categorical if users set the numeric_categorical_threshold config variable to specify the number of unique values that should be used as a cutoff.

We might want to update this default behavior so that numeric columns that contain a small percentage of unique values relative to the number of data points are inferred as Categorical. This could also be accompanied by changing the threshold variable to be a percentage instead of a raw number of unique values.