Rail optimization for rings (#1140)

- Modifies the ring creation algorithm to be friendlier to rail-optimized topologies (should not affect classic fabric topologies)

[ROCm/rccl commit: 4cb62f999a]

projects/rccl/tools/topo_expl/topo_expl.cpp

@@ -158,6 +158,7 @@ NodeModelDesc model_descs[] = {
   {2, "topo_8p_940vm.xml",      "2 nodes gfx940 VM"},
   {2, "topo_8p_940_16n.xml",    "2 nodes gfx940 16 NICs"},
   {2, "topo_8p1h_6.xml",        "2 nodes 8P1H Alt."},
+  {5, "topo_8p_940.xml",        "5 nodes gfx940 8P"},
 };
 
 NCCL_PARAM(MaxCTAs, "MAX_CTAS", MAXCHANNELS);