Remove MPI operations inside of decompInit_glcp

Remove the MPI operations inside of decompInit_glcp and do it only over the local processor. This adds unneeded complexity, and a small amount of global non-scalable memory (only on masterproc dimensions both by total global gridcells and total number of processors). Since it's only on masterproc this is only a savings of a a few MBytes. But, the complexity and the additional overhead in MPI buffers might be worthwhile. The timers show the time spent is essentially zero, so it won't matter for execution time.