[SPARK-19816][SQL][TESTS] Fix an issue that DataFrameCallbackSuite doesn't recover the log level

## What changes were proposed in this pull request?

"DataFrameCallbackSuite.execute callback functions when a DataFrame action failed" sets the log level to "fatal" but doesn't recover it. Hence, tests running after it won't output any logs except fatal logs.

This PR uses `testQuietly` instead to avoid changing the log level.

## How was this patch tested?

Jenkins

Author: Shixiong Zhu <shixiong@databricks.com>

Closes #17156 from zsxwing/SPARK-19816.

sql/core/src/test/scala/org/apache/spark/sql/util/DataFrameCallbackSuite.scala

@@ -60,7 +60,7 @@ class DataFrameCallbackSuite extends QueryTest with SharedSQLContext {
     spark.listenerManager.unregister(listener)
   }
 
-  test("execute callback functions when a DataFrame action failed") {
+  testQuietly("execute callback functions when a DataFrame action failed") {
     val metrics = ArrayBuffer.empty[(String, QueryExecution, Exception)]
     val listener = new QueryExecutionListener {
       override def onFailure(funcName: String, qe: QueryExecution, exception: Exception): Unit = {
@@ -75,8 +75,6 @@ class DataFrameCallbackSuite extends QueryTest with SharedSQLContext {
     val errorUdf = udf[Int, Int] { _ => throw new RuntimeException("udf error") }
     val df = sparkContext.makeRDD(Seq(1 -> "a")).toDF("i", "j")
 
-    // Ignore the log when we are expecting an exception.
-    sparkContext.setLogLevel("FATAL")
     val e = intercept[SparkException](df.select(errorUdf($"i")).collect())
 
     assert(metrics.length == 1)