Prevent TPOT from creating multiple copies of the data set (except at leaf nodes)

[rhiever]

In many cases, TPOT creates multiple copies of the data set when it runs through multiple pipeline operators. The only time TPOT should have to store multiple copies of the data set is when multiple leaf nodes introduce multiple copies. Make sure TPOT cleans up data set copies as it processes them.

[rhiever]

This will be in the 0.5 release because we work directly with sklearn Pipelines now.