RDD zipWithIndex generate wrong result when one partition contains more than Int.MaxValue records.
when RDD contains a partition with more than 2147483647 records,
for example, if partition-0 has more than 2147483647 records, the index became:
0,1, ..., 2147483647, -2147483648, -2147483647, -2147483646 ....
when we do some operation such as repartition or coalesce, it is possible to generate big partition, so this bug should be fixed.