[SPARK-24172][SQL] we should not apply operator pushdown to data source v2 many times

## What changes were proposed in this pull request?

In `PushDownOperatorsToDataSource`, we use `transformUp` to match `PhysicalOperation` and apply pushdown. This is problematic if we have multiple `Filter` and `Project` above the data source v2 relation.

e.g. for a query
```
Project
  Filter
    DataSourceV2Relation
```

The pattern match will be triggered twice and we will do operator pushdown twice. This is unnecessary, we can use `mapChildren` to only apply pushdown once.

## How was this patch tested?

existing test

Author: Wenchen Fan <wenchen@databricks.com>

Closes #21230 from cloud-fan/step2.

Author: cloud-fan

sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/v2/PushDownOperatorsToDataSource.scala

@@ -23,17 +23,10 @@ import org.apache.spark.sql.catalyst.plans.logical.{Filter, LogicalPlan, Project
 import org.apache.spark.sql.catalyst.rules.Rule
 
 object PushDownOperatorsToDataSource extends Rule[LogicalPlan] {
-  override def apply(
-      plan: LogicalPlan): LogicalPlan = plan transformUp {
+  override def apply(plan: LogicalPlan): LogicalPlan = plan match {
     // PhysicalOperation guarantees that filters are deterministic; no need to check
-    case PhysicalOperation(project, newFilters, relation : DataSourceV2Relation) =>
-      // merge the filters
-      val filters = relation.filters match {
-        case Some(existing) =>
-          existing ++ newFilters
-        case _ =>
-          newFilters
-      }
+    case PhysicalOperation(project, filters, relation: DataSourceV2Relation) =>
+      assert(relation.filters.isEmpty, "data source v2 should do push down only once.")
 
       val projectAttrs = project.map(_.toAttribute)
       val projectSet = AttributeSet(project.flatMap(_.references))
@@ -67,5 +60,7 @@ object PushDownOperatorsToDataSource extends Rule[LogicalPlan] {
       } else {
         filtered
       }
+
+    case other => other.mapChildren(apply)
   }
 }