Support map_filter operator

[res-life]

Closes #5273

Support map_filter operator.

map_filter is used to filter entries in a map.

The usage of map_filter is:

map_filter(map_column_expession, (key, value) -> prediate(key, value))

e.g.:

>>> df = spark.sql("SELECT map_from_arrays(array(1, 2, 3), array(1, 2, 3)) as map")
>>> df.show(truncate = False)
+------------------------+
|map                     |
+------------------------+
|{1 -> 1, 2 -> 2, 3 -> 3}|
+------------------------+

>>> df.selectExpr("map_filter(map, (key, value) -> key + value > 5) as ret").show(truncate = False)
+--------+
|ret     |
+--------+
|{3 -> 3}|
+--------+

This implementation is:

1. apply the predicate function in map_filter to get a plain boolean column
2. convert the plain boolean column to a list-of-boolean column according to the offsets column of inputted map column
3. invoke cuDF apply_boolean_mask method to filter, Refer to Segmented apply_boolean_mask for LIST columns rapidsai/cudf#10773

Signed-off-by: Chong Gao res_life@163.com

[res-life]

Depending on JNI PR rapidsai/cudf#10812 to merge.

[firestarman]

Looks good to me

[firestarman]

NIT:

Suggested change

	override def dataType: DataType = MapType(keyType, valueType, valueContainsNull)
	override def dataType: DataType = argument.dataType

[res-life]

Done.

[revans2]

Agree with the one nit by @firestarman looks really good.

[res-life]

build