[BUG] Implement hasSideEffects for all expressions that have side-effects

[andygrove]

Describe the bug
In #3849 we implemented special handling in CASE WHEN for expressions with side-effects but this relies on all expressions with side-effects overriding the hasSideEffects method and we have not done this for all expressions (see #5023 for one example).

We should audit all expressions and implement hasSideEffects where appropriate.

Candidates:

• GpuGetArrayItem -
• GpuElementAt
• GpuGetMapValue
• GpuUnaryMinus
• GpuAbs
• GpuSum

Steps/Code to reproduce bug
N/A

Expected behavior
Conditional expressions such as IF and CASE should not fail due to expressions with side-effects being evaluated and throwing exceptions.

Environment details (please complete the following information)
N/A

Additional context
N/A

[revans2]

I just did a test with pure spark 3.2.0 and I can confirm that case/when does not stop a SUM in ansi from throwing an exception if the case/when is outside of the SUM.

ANSI DISABLED:

scala>  Seq((0L, Long.MaxValue), (0L, 100L), (1L, 100L)).toDF.groupBy("_1").agg(when(col("_1") === 1, sum("_2")).otherwise(0).alias("result")).show()
+---+------+
| _1|result|
+---+------+
|  0|     0|
|  1|   100|
+---+------+

ANSI ENABLED:

scala>  Seq((0L, Long.MaxValue), (0L, 100L), (1L, 100L)).toDF.groupBy("_1").agg(when(col("_1") === 1, sum("_2")).otherwise(0).alias("result")).show()
22/04/15 19:08:34 ERROR Executor: Exception in task 0.0 in stage 10.0 (TID 29)
java.lang.ArithmeticException: long overflow
	at java.lang.Math.addExact(Math.java:809)
	at org.apache.spark.sql.catalyst.expressions.GeneratedClass$GeneratedIteratorForCodegenStage1.agg_doConsume_0$(Unknown Source)
...

If I move the case/when inside the SUM, then it does the right thing in both places...

scala>  Seq((0L, Long.MaxValue), (0L, 100L), (1L, 100L)).toDF.groupBy("_1").agg(sum(when(col("_1") === 1, col("_2")).otherwise(0)).alias("result")).show()
+---+------+
| _1|result|
+---+------+
|  0|     0|
|  1|   100|
+---+------+