Support a variable number of columns in the array feature selection transform

[riley-harper]

This PR closes #134.

Previously, the array feature selection transform defined in hlink/linking/core/transforms.py required exactly two input columns. This change updates the transform to accept any number of columns instead. This is a small change because the pyspark.sql.functions.array() function already accepts a variable number of input columns to pack into the output column.

I also took this opportunity to add a few tests specifically for this feature and add type hints to the generate_transforms() function.

[joegrover]

Looks good to me. I just had one question which I attached to the specific line.

[joegrover]

I don't really know how the pyspark.sql.functions.array() works when a list is passed, though your tests seem to indicate it's fine, but this makes it look like you can just pass the list itself as an "array literal". Or does passing it this way allow for the "input_columns" to be either a single string or a list of strings?

[riley-harper]

I didn't think to look too closely at that either since the tests passed.

Pyspark automatically flattens the argument when you pass a single list to array(): https://api-docs.databricks.com/python/pyspark/latest/pyspark.sql/api/pyspark.sql.functions.array.html. So array(input_cols) and array(*input_cols) are equivalent here. I guess that passing it this way allows the input_columns in the config to be either a string or list! But that was not the intention of doing it this way.

I think the "array literal" syntax might be out of date, because I get a type error from it.

from hlink.spark.factory import SparkFactory

factory = SparkFactory()
spark = factory.create()
df = spark.createDataFrame([[1, 2, 3], [2, 3, 4]], schema=["A", "B", "C"])
df2 = df.withColumn("D", ["A", "B"])

gives me

pyspark.errors.exceptions.base.PySparkTypeError: [NOT_COLUMN] Argument `col` should be a Column, got list.