IntelMPI bug with Slurm jobs

[anhoward]

For some reason, it looks like Intel MPI jobs running on Slurm don't work properly if you don't set

I_MPI_HYDRA_IFACE=eth0

The initial job on a cluster works, but all subsequent jobs fail. Still trying to track down the actual cause, but if we can set this in the impi module that would be great.