[FEA] Support ScalarSubquery

[sperlingxx]

Is your feature request related to a problem? Please describe.
ScalarSubquery, which turning the output of a (sub)query into a single scalar value, has yet supported under GPU. This feature is important for better performance.

Describe the solution you'd like
Implement GPU overrides for org.apack.spark.sql.execution.SubqueryExec and org.apack.spark.sql.execution.ScalarSubquery.

[revans2]

I am not sure we need to implement SubqueyExec for this use case. ScalarSubquery will run a query that produces a single result. One row of one column. It will run that sub-query and collect the result back to the driver.

https://github.com/apache/spark/blob/5718d64f3104f7a24a9d4b619748bcca03031c48/sql/core/src/main/scala/org/apache/spark/sql/execution/subquery.scala#L82-L96

But then it keeps the data cached in the expression and just returns that value as if it were a literal.

There is really almost no value at all in trying to make the various SubqueryExecBase implementations run on the GPU because it is returning just a single value. If it was more data then it might make since to do something like BroadcastExchangeExec so we can keep a large amount of the data columnar. Here the amount of data is likely small enough that there is no point.