Support GpuMapConcat

[HaoYang670]

Signed-off-by: remzi [email protected]
Closes #5405.

Method of concatenating map

1. concatenate the key child and value child
2. Build new map column vector from the above.

Supported types

MapType(primitive types) can run on GPU.
Nested types (MapType(array, struct, map)) are not supported. Because ColumnVector.listConcatenateByRow doesn't support nested types.

Example

scala> val df = Seq((Map(1 -> 2, 9 -> 10), Map(5 -> 6)), (Map(3 -> 4), Map(7 -> 8, 11 -> 12))).toDF
df: org.apache.spark.sql.DataFrame = [_1: map<int,int>, _2: map<int,int>]

scala> df.show
22/05/19 11:55:29 WARN GpuOverrides: 
! <LocalTableScanExec> cannot run on GPU because GPU does not currently support the operator class org.apache.spark.sql.execution.LocalTableScanExec
  @Expression <AttributeReference> _1#20 could run on GPU
  @Expression <AttributeReference> _2#21 could run on GPU

+-----------------+------------------+
|               _1|                _2|
+-----------------+------------------+
|{1 -> 2, 9 -> 10}|          {5 -> 6}|
|         {3 -> 4}|{7 -> 8, 11 -> 12}|
+-----------------+------------------+

scala> df.select(map_concat($"_1", $"_2")).collect
22/05/19 11:56:06 WARN GpuOverrides: 
! <LocalTableScanExec> cannot run on GPU because GPU does not currently support the operator class org.apache.spark.sql.execution.LocalTableScanExec
  @Expression <AttributeReference> map_concat(_1, _2)#34 could run on GPU

res2: Array[org.apache.spark.sql.Row] = Array([Map(1 -> 2, 9 -> 10, 5 -> 6)], [Map(3 -> 4, 7 -> 8, 11 -> 12)])

[HaoYang670]

build

[HaoYang670]

build

[revans2]

Just a few nits that I don't think would block this from going in.

[HaoYang670]

build

[HaoYang670]

build

[HaoYang670]

build

[HaoYang670]

Hi @revans2 I think I have addressed all of your comments. Please review, thank you!