Add pthread_detach()'s for threads we never pthread_join(). Helps

reduce diagnostic noise for ThreadSanitizer.

Fixes https://github.com/NVIDIA/nccl/issues/649

src/bootstrap.cc

@@ -165,6 +165,7 @@ ncclResult_t bootstrapCreateRoot(ncclUniqueId* id, bool idFromEnv) {
   memcpy(id, &listenSock->addr, sizeof(union ncclSocketAddress));
   pthread_t thread;
   pthread_create(&thread, NULL, bootstrapRoot, (void*)listenSock);
+  pthread_detach(thread); // will not be pthread_join()'d
   ncclSetThreadName(thread, "NCCL BootstrapR");
   return ncclSuccess;
 }

src/proxy.cc

@@ -1073,6 +1073,7 @@ ncclResult_t ncclProxyInit(struct ncclComm* comm, struct ncclSocket* sock, union
 }
 
 ncclResult_t ncclProxyCreate(struct ncclComm* comm) {
+  // comm->proxyState.thread is pthread_join()'d by commFree() in init.cc
   pthread_create(&comm->proxyState.thread, NULL, ncclProxyService, comm);
   return ncclSuccess;
 }

src/transport/net_ib.cc

@@ -223,6 +223,7 @@ ncclResult_t ncclIbInit(ncclDebugLogger_t logFunction) {
           ncclIbDevs[ncclNIbDevs].mrCache.slots = NULL;
 
           pthread_create(&ncclIbAsyncThread, NULL, ncclIbAsyncThreadMain, context);
+          pthread_detach(ncclIbAsyncThread); // will not be pthread_join()'d
           ncclSetThreadName(ncclIbAsyncThread, "NCCL IbAsync %2d", ncclNIbDevs);
           ncclNIbDevs++;
           nPorts++;