variance reduction with cell importances

[eepeterson]

the fns_duct MCNP input uses cell importances for neutrons. We could (and should) get around this by generating weight windows with Attila and using those instead, but I just wanted to track the fact that this would be different between the benchmarks.