fix(spark): nullability of output columns of expand relation (#310)

Fix the logic for deriving whether each output column of the
expand relation is nullable or not.
If one or more of the projections can assign null to the column,
then it must be defined as nullable.

Signed-off-by: Andrew Coleman <andrew_coleman@uk.ibm.com>

spark/src/main/scala/io/substrait/spark/logical/ToLogicalPlan.scala

@@ -224,9 +224,12 @@ class ToLogicalPlan(spark: SparkSession) extends DefaultRelVisitor[LogicalPlan]
             throw new UnsupportedOperationException("ConsistentField not currently supported")
         }
 
-      val output = projections.head
+      // An output column is nullable if any of the projections can assign null to it
+      val types = projections.transpose.map(p => (p.head.dataType, p.exists(_.nullable)))
+
+      val output = types
         .zip(names)
-        .map { case (t, name) => StructField(name, t.dataType, t.nullable) }
+        .map { case (t, name) => StructField(name, t._1, t._2) }
         .map(f => AttributeReference(f.name, f.dataType, f.nullable, f.metadata)())
 
       Expand(projections, output, child)

spark/src/test/scala/io/substrait/spark/TPCDSPlan.scala

@@ -33,7 +33,7 @@ class TPCDSPlan extends TPCDSBase with SubstraitPlanTestBase {
 
   // spotless:off
   val successfulSQL: Set[String] = Set("q1", "q3", "q4", "q7",
-    "q11", "q13", "q14b", "q15", "q16", "q18", "q19",
+    "q11", "q13", "q14a", "q14b", "q15", "q16", "q18", "q19",
     "q21", "q22", "q23a", "q23b", "q24a", "q24b", "q25", "q26", "q28", "q29",
     "q30", "q31", "q32", "q33", "q37", "q38",
     "q40", "q41", "q42", "q43", "q46", "q48",