Enable MSCCLPP use in CPX mode (#1355)

This PR enables the use of MSCCLPP in CPX mode for 8 GPUs.

ext-src/cpx.patch

@@ -0,0 +1,12 @@
+diff --git a/src/numa.cc b/src/numa.cc
+index d72c99e..16c903d 100644
+--- a/src/numa.cc
++++ b/src/numa.cc
+@@ -26,6 +26,7 @@ namespace mscclpp {
+ 
+ MSCCLPP_API_CPP int getDeviceNumaNode(int cudaDev) {
+   std::string busId = getBusId(cudaDev);
++  busId[busId.length() - 1] = '0';
+   std::string file_str = "/sys/bus/pci/devices/" + busId + "/numa_node";
+   std::ifstream file(file_str);
+   int numaNode;