Add support for external network.

Dynamically load external network from libnccl-net.so.
Add init function in networks.
Move PCI scoring to net.cu, only ask transport to provide a path.
Simplify CUDA PCI path detection.
Add dummy external network

src/misc/group.cu

@@ -58,7 +58,7 @@ ncclResult_t ncclSetDevice(int cudaDev) {
 
 #define CHECK(a) do { \
   if ((args->ret = (a)) != ncclSuccess) { \
-    INFO(INIT,"%s:%d -> %d [Async thread]", __FILE__, __LINE__, args->ret); \
+    INFO(NCCL_INIT,"%s:%d -> %d [Async thread]", __FILE__, __LINE__, args->ret); \
     return args; \
   } \
 } while(0)