Implement duplicate key handling for GpuCreateMap

[andygrove]

Signed-off-by: Andy Grove andygrove@nvidia.com

Closes #3250

Depends on rapidsai/cudf#9553

scala> val df = Seq(1,2,3).toDF("c0")
df: org.apache.spark.sql.DataFrame = [c0: int]

scala> df.repartition(2).createOrReplaceTempView("foo")

scala> spark.sql("SELECT map(c0, 1, 1 + c0 - 1, 2) FROM foo").collect

21/11/02 20:46:05 ERROR Executor: Exception in task 0.0 in stage 2.0 (TID 3)
java.lang.RuntimeException: Duplicate map key was found, please check the input data. If you want to remove the duplicated keys, you can set spark.sql.mapKeyDedupPolicy to LAST_WIN so that the key inserted at last takes precedence.
	at com.nvidia.spark.rapids.GpuMapUtils$.duplicateMapKeyFoundError(GpuMapUtils.scala:134)
	at org.apache.spark.sql.rapids.GpuCreateMap.$anonfun$columnarEval$10(complexTypeCreator.scala:118)
	
scala> spark.conf.set("spark.sql.mapKeyDedupPolicy", "LAST_WIN")

scala> spark.sql("SELECT map(c0, 1, 1 + c0 - 1, 2) FROM foo").collect

res3: Array[org.apache.spark.sql.Row] = Array([Map(1 -> 2)], [Map(2 -> 2)], [Map(3 -> 2)])

[andygrove]

build

[jlowe]

build

[andygrove]

build

[revans2]

build