Test problem added to address bug 6339.

With OBJ_WEIGHT_DIM=2, Tom Hughes reported different results using RCB
than using HIER with RCB on sixteen 16-core nodes.  
In truth, the partitions should be nearly identical.

This test problem replicates the main features of Tom's problem.
It has a uniform 20x20x20 mesh.  Particles are distributed in the mesh
such that they are injected at one end in a small radius.  The number of
particles per cell within the radius decreases for cells further from the
injection point, and is zero within some distance opposite the injection point.
See hughes.20x20x20.stats for details.

The Chaco files were generated with gen2.py.  The calling sequence is in
hughes.20x20x20.stats.

This test exercises the HIER and RCB methods with two weights; the answers
should be nearly identical.  It also exercises "test local partitions = 9"
which replicates the conditions that caused the error in Tom's tests.


