Add support for structs in GpuScalarSubquery

[tgravescs]

Is your feature request related to a problem? Please describe.

Recently I found a case where a query was optimized so that the hash aggregate returned multiple aggregates in a struct, but our GpuScalarSubquery doesn't support handling of structs. We should add support for this.

 :  :- Subquery scalar-subquery#56, [id=#54]
  :  :  +- *(2) Project [named_struct(count(s), count(s)#31L, max(d), max(d)#33, max(i), max(i)#17, min(i), min(i)#19) AS mergedValue#55]
  :  :     +- *(2) HashAggregate(keys=[], functions=[count(s#37), max(d#38), max(i#35), min(i#35)])
  :  :        +- Exchange SinglePartition, ENSURE_REQUIREMENTS, [id=#49]
  :  :           +- *(1) HashAggregate(keys=[], functions=[partial_count(s#37), partial_max(d#38), partial_max(i#35), partial_min(i#35)])
  :  :              +- *(1) Project [s#37, d#38, i#35]
  :  :                 +- *(1) Scan ExistingRDD[l#34L,i#35,f#36,s#37,d#38]

!Expression <ScalarSubquery> Subquery scalar-subquery#150, [id=#300] cannot run on GPU because expression ScalarSubquery Subquery scalar-subquery#150, [id=#300] produces an unsupported type StructType(StructField(count(s),LongType,false), StructField(max(d),DecimalType(38,10),true), StructField(max(i),IntegerType,true), StructField(min(i),IntegerType,true))

Specifically in this environment the test_scalar_subquery test failed because the Project wasn't on the GPU.

[tgravescs]

note that this type of optimization in spark could be reproduced by a change like: apache/spark#32298