[FEA] Support approx_percentile in reduction context

[viadea]

I wish approx_percentile can be supported in reduction context.
For example:

spark.conf.set("spark.rapids.sql.incompatibleOps.enabled",true)
val data = Seq((1,"abc.com"),(2,"...abc"),(3,".a.b.c"))
import spark.implicits._
val df = data.toDF("id","txt")
df.write.mode("overwrite").format("parquet").save("/tmp/testparquet")
val df2=spark.read.parquet("/tmp/testparquet")
df2.createOrReplaceTempView("df2")
spark.sql("select approx_percentile(id,0.75) from df2").show()

Not-supported-messages:

    !Exec <ObjectHashAggregateExec> cannot run on GPU because not all expressions can be replaced. The data type of following expressions will be converted in GPU runtime: buf#65: Converted BinaryType to StructType(StructField(centroids,ArrayType(StructType(StructField(mean,DoubleType,false), StructField(weight,DoubleType,false)),false),true), StructField(min,DoubleType,false), StructField(max,DoubleType,false))
      @Expression <AggregateExpression> partial_approx_percentile(id#54, 0.75, 10000, 0, 0) AS buf#65 could run on GPU
        !Expression <ApproximatePercentile> approx_percentile(id#54, 0.75, 10000, 0, 0) cannot run on GPU because accuracy expression Literal 10000 (IntegerType is not supported); percentage expression Literal 0.75 (DoubleType is not supported); expression ApproximatePercentile approx_percentile(id#54, 0.75, 10000, 0, 0) produces an unsupported type IntegerType; input expression AttributeReference id#54 (IntegerType is not supported)

[ttnghia]

This depends on cudf, which in turn requires additional implementation. Reduction and groupby can share common code but not the same.