[SPARK-15392][SQL] fix default value of size estimation of logical plan

## What changes were proposed in this pull request?

We use  autoBroadcastJoinThreshold + 1L as the default value of size estimation, that is not good in 2.0, because we will calculate the size based on size of schema, then the estimation could be less than autoBroadcastJoinThreshold if you have an SELECT on top of an DataFrame created from RDD.

This PR change the default value to Long.MaxValue.

## How was this patch tested?

Added regression tests.

Author: Davies Liu <davies@databricks.com>

Closes #13179 from davies/fix_default_size.

sql/core/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala

@@ -605,8 +605,7 @@ private[sql] class SQLConf extends Serializable with CatalystConf with Logging {
 
   def enableRadixSort: Boolean = getConf(RADIX_SORT_ENABLED)
 
-  def defaultSizeInBytes: Long =
-    getConf(DEFAULT_SIZE_IN_BYTES, autoBroadcastJoinThreshold + 1L)
+  def defaultSizeInBytes: Long = getConf(DEFAULT_SIZE_IN_BYTES, Long.MaxValue)
 
   def isParquetBinaryAsString: Boolean = getConf(PARQUET_BINARY_AS_STRING)
 

sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala

@@ -1476,4 +1476,13 @@ class DataFrameSuite extends QueryTest with SharedSQLContext {
       getMessage()
     assert(e1.startsWith("Path does not exist"))
   }
+
+  test("SPARK-15392: DataFrame created from RDD should not be broadcasted") {
+    val rdd = sparkContext.range(1, 100).map(i => Row(i, i))
+    val df = spark.createDataFrame(rdd, new StructType().add("a", LongType).add("b", LongType))
+    assert(df.queryExecution.analyzed.statistics.sizeInBytes >
+      spark.wrapped.conf.autoBroadcastJoinThreshold)
+    assert(df.selectExpr("a").queryExecution.analyzed.statistics.sizeInBytes >
+      spark.wrapped.conf.autoBroadcastJoinThreshold)
+  }
 }