Rail optimized trees (#1540)

* Allow disabling rail-optimized trees via RCCL_DISABLE_RAIL_TREES, Graphviz-friendly output via RCCL_OUTPUT_TREES

src/graph/topo.h

@@ -192,6 +192,7 @@ struct ncclTopoSystem {
 
   // [RCCL] Track hostIdx to support rail-optimized rings/trees
   int hostIdx;
+  bool useRailOptimizedTrees;
 };
 
 ncclResult_t ncclTopoGetNode(struct ncclTopoSystem* system, struct ncclTopoNode** node, int type, uint64_t id);