Make worker->shuttingdown atomic

This record member could be accessed concurrently from within the
context of multiple threads. For example, it is used in
isc__nmsocket_closing(), which can be called on listener sockets from
a context of a thread which does not own the object.

This lead to TSAN reports, this commit should fix that.

lib/isc/netmgr/netmgr-int.h

@@ -202,7 +202,7 @@ typedef struct isc__networker {
 	isc_refcount_t references;
 	isc_loop_t *loop;
 	isc_nm_t *netmgr;
-	bool shuttingdown;
+	atomic_bool shuttingdown;
 
 	char *recvbuf;
 	char *sendbuf;

lib/isc/netmgr/netmgr.c

@@ -156,7 +156,7 @@ networker_teardown(void *arg) {
 	isc__networker_t *worker = arg;
 	isc_loop_t *loop = worker->loop;
 
-	worker->shuttingdown = true;
+	atomic_store(&worker->shuttingdown, true);
 
 	isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL, ISC_LOGMODULE_NETMGR,
 		      ISC_LOG_DEBUG(1),
@@ -1705,7 +1705,7 @@ isc__nm_stop_reading(isc_nmsocket_t *sock) {
 
 bool
 isc__nm_closing(isc__networker_t *worker) {
-	return (worker->shuttingdown);
+	return (atomic_load(&worker->shuttingdown));
 }
 
 bool