[FEA] Support max on single-level struct in aggregation context

[nvliyuan]

Is your feature request related to a problem? Please describe.
I wish the plugin can support Max on Single-struct in aggregation context.

Additional context
Code example:

import org.apache.spark.sql.Row
import org.apache.spark.sql.types._

val data = Seq(
    Row(Row("Adam ","","Green"),1,"M",10,"Math:13,Gym:24,English:45",Map("hair"->"black","eye"->"black"),"{\"hair\":\"black\",\"eye\":\"black\"}"),
    Row(Row("Bob ","Middle","Green"),2,"M",20,"Math:55,Gym:24,English:37",Map("hair"->"yellow","eye"->"yellow"),"{\"hair\":\"green\",\"eye\":\"green\"}"),
    Row(Row("Cathy ","","Green"),3,"F",30,"Math:83,Gym:15,English:63",Map("hair"->"blue","eye"->"blue"),"{\"hair\":\"blue\",\"eye\":\"blue\"}")
)

val schema = (new StructType()
  .add("name",new StructType()
    .add("firstname",StringType)
    .add("middlename",StringType)
    .add("lastname",StringType)) 
  .add("low",IntegerType)
  .add("gender",StringType)
  .add("high",IntegerType)
  .add("score",StringType)
  .add("feature",MapType(StringType,StringType))
  .add("feature_json",StringType))

val df = spark.createDataFrame(spark.sparkContext.parallelize(data),schema)
df.write.format("parquet").mode("overwrite").save("/tmp/yl/tmpdatas")

val df2 = spark.read.parquet("/tmp/yl/tmpdatas")
df2.createOrReplaceTempView("df")


spark.sql("SELECT max(name) from df").show()

[sameerz]

Depends on rapidsai/cudf#8974

[ttnghia]

The example above is a reduction context. Currently on track of development, targetting for 22.02 release.