spark tgz-s need to be deleted on mvn clean, as are other binary artifacts in the tree

Currently, tgz files are downloaded and kept around forever. I noticed when packaging the release (apparently the excludes in packaging files also didn't work) that the initial src tar.gz was huge; regardless of that, I had 6 version of spark (1.2 thru 1.6 with one dot version) sitting there, and also in every clone of Hive that I have.

These should be switched to use normal means of artifact distribution (I think I already filed a jira but I cannot find it now); meanwhile making sure that mvn clean would remove them.

I realize it could create some pain when running tests repeatedly on dev machine unless "clean" is omitted from rebuilds; that is somewhat intentional - it should be a good incentive to switch to maven for dependency management instead of a bash script