[SPARK-6082] [SQL] Provides better error message for malformed rows when caching tables

Constructs like Hive `TRANSFORM` may generate malformed rows (via badly authored external scripts for example). I'm a bit hesitant to have this feature, since it introduces per-tuple cost when caching tables. However, considering caching tables is usually a one-time cost, this is probably worth having.

<!-- Reviewable:start -->
[<img src="https://reviewable.io/review_button.png" height=40 alt="Review on Reviewable"/>](https://reviewable.io/reviews/apache/spark/4842)
<!-- Reviewable:end -->

Author: Cheng Lian <lian@databricks.com>

Closes #4842 from liancheng/spark-6082 and squashes the following commits:

b05dbff [Cheng Lian] Provides better error message for malformed rows when caching tables

sql/core/src/main/scala/org/apache/spark/sql/columnar/InMemoryColumnarTableScan.scala

@@ -119,6 +119,17 @@ private[sql] case class InMemoryRelation(
           var rowCount = 0
           while (rowIterator.hasNext && rowCount < batchSize) {
             val row = rowIterator.next()
+
+            // Added for SPARK-6082. This assertion can be useful for scenarios when something
+            // like Hive TRANSFORM is used. The external data generation script used in TRANSFORM
+            // may result malformed rows, causing ArrayIndexOutOfBoundsException, which is somewhat
+            // hard to decipher.
+            assert(
+              row.size == columnBuilders.size,
+              s"""Row column number mismatch, expected ${output.size} columns, but got ${row.size}.
+                 |Row content: $row
+               """.stripMargin)
+
             var i = 0
             while (i < row.length) {
               columnBuilders(i).appendFrom(row, i)