MPI-PR: incorrect detection of /dev/shm usage when GA_NUM_PROGRESS_RANKS_PER_NODE>1

[edoapra]

Since the code uses g_state.node_size -1 to estimate the compute process on a give node,
when GA_NUM_PROGRESS_RANKS_PER_NODE>1 this formula is wrong and should become
(g_state.node_size -1)*GA_NUM_PROGRESS_RANKS_PER_NODE, instead.