[BUG] ExpressionEncoder does not work for input in GpuScalaUDF

[firestarman]

Describe the bug
It complains the below errors when enabling ExpressionEncoder to do conversions for inputs of an UDF.

  Cause: java.lang.RuntimeException: Error while decoding: org.apache.spark.sql.catalyst.analysis.UnresolvedException: Invalid call to nullable on unresolved object, tree: getcolumnbyordinal(0, ShortType)
staticinvoke(class java.lang.Short, ObjectType(class java.lang.Short), valueOf, upcast(getcolumnbyordinal(0, ShortType), ShortType, - root class: "java.lang.Short"), true, false)
  at org.apache.spark.sql.catalyst.encoders.ExpressionEncoder$Deserializer.apply(ExpressionEncoder.scala:188)
  at org.apache.spark.sql.rapids.GpuScalaUDF.$anonfun$scalaConverter$2(GpuScalaUDF.scala:108)
  at org.apache.spark.sql.rapids.GpuScalaUDF.$anonfun$childAccessors$2(GpuScalaUDF.scala:141)
  at org.apache.spark.sql.rapids.GpuScalaUDF.$anonfun$argsParser$2(GpuScalaUDF.scala:145)
  at org.apache.spark.sql.rapids.GpuScalaUDF.$anonfun$argsParser$2$adapted(GpuScalaUDF.scala:145)
  at scala.collection.TraversableLike.$anonfun$map$1(TraversableLike.scala:237)
  at scala.collection.immutable.Range.foreach(Range.scala:158)
  at scala.collection.TraversableLike.map(TraversableLike.scala:237)
  at scala.collection.TraversableLike.map$(TraversableLike.scala:230)
  at scala.collection.AbstractTraversable.map(Traversable.scala:108)
  ...
  Cause: org.apache.spark.sql.catalyst.analysis.UnresolvedException: Invalid call to nullable on unresolved object, tree: null
  at org.apache.spark.sql.catalyst.analysis.GetColumnByOrdinal.nullable(unresolved.scala:520)
  at org.apache.spark.sql.catalyst.expressions.UnaryExpression.nullable(Expression.scala:450)
  at org.apache.spark.sql.catalyst.expressions.objects.InvokeLike.$anonfun$needNullCheck$1(objects.scala:48)
  at org.apache.spark.sql.catalyst.expressions.objects.InvokeLike.$anonfun$needNullCheck$1$adapted(objects.scala:48)
  at scala.collection.LinearSeqOptimized.exists(LinearSeqOptimized.scala:95)
  at scala.collection.LinearSeqOptimized.exists$(LinearSeqOptimized.scala:92)
  at scala.collection.immutable.List.exists(List.scala:89)
  at org.apache.spark.sql.catalyst.expressions.objects.InvokeLike.needNullCheck(objects.scala:48)
  at org.apache.spark.sql.catalyst.expressions.objects.InvokeLike.needNullCheck$(objects.scala:48)
  at org.apache.spark.sql.catalyst.expressions.objects.StaticInvoke.needNullCheck$lzycompute(objects.scala:211)

This is a follow on issue for #3897 .

Steps/Code to reproduce bug
Enable the encoder and run the tests as below
mvn clean test -nsu -DwildcardSuites="com.nvidia.spark.rapids.CPUBasedUDFSuite"

This is low priority since we can use CatalystTypeConverters to do the same things, and it is good enough for cudf supported types.

[firestarman]

This is low priority since we can use CatalystTypeConverters to do the same things, and it is good enough for cudf supported types. Added P2 tag.

[firestarman]

Here is a related issue #3855

[firestarman]

The code for input type conversion is changed a lot from Spark 3.0 to 3.1+.

This error happened when running the 3.1+ code under Spark 3.0. So it dismisses after changing to use the 3.0 code now.

Close this issue.