Subject: -- misplaced calls to setJarByClass

Current hadoop has a mechanism for storing dependencies into hdfs. We
don't use it. We use this self-contained jar thing.

If I read this right, we might be better off if we didn't call
'setJarByClass' at all.

By default, I think that hadoop assumes that the jar passed on the
command line is the right one to pass around. We mess this up by
calling setJarByClass.

Experimental results on this theory will be forthcoming very soon.