[SPARK-10540] Fixes flaky all-data-type test

This PR breaks the original test case into multiple ones (one test case for each data type). In this way, test failure output can be much more readable.

Within each test case, we build a table with two columns, one of them is for the data type to test, the other is an "index" column, which is used to sort the DataFrame and workaround [SPARK-10591] [1]

[1]: https://issues.apache.org/jira/browse/SPARK-10591

Author: Cheng Lian <lian@databricks.com>

Closes #8768 from liancheng/spark-10540/test-all-data-types.