Fix documentation for DataSetUtils.zipWithUniqueId()

Under FLINK-2590 the assignment and testing of unique IDs was improved but the documentation looks to still reference the old implementation.

With parallelism=1 there is no difference between zipWithUniqueID and zipWithIndex. With greater parallelism the results of zipWithUniqueID are dependent on the partitioning.

The documentation should demonstrate a possible result that is different from the incremental sequence of zipWithIndex while noting that results are dependent on the parallelism and partitioning.