Jenkins should compile Spark against multiple versions of Hadoop

The build broke because of PR https://github.com/apache/spark/pull/2609#issuecomment-57962393 - however the build failure was not caught by Jenkins. From what I understand the build failure occurs when Spark is built manually against certain versions of Hadoop.

It seems intuitive that Jenkins should catch this sort of thing. The code should be compiled against multiple Hadoop versions. It seems like overkill to run the full test suite against all Hadoop versions.