cluster: use start_single where appropriate

For these two classes, it became noticable when adding
shutdown logging that they were instantiated on all cores,
whereas only used on shard 0.

Using start_single saves us a few bytes of RAM and expresses
the intent more clearly.

src/v/cluster/controller.cc

@@ -137,7 +137,7 @@ ss::future<> controller::start() {
             std::ref(_as));
       })
       .then([this] {
-          return _config_manager.start(
+          return _config_manager.start_single(
             std::ref(_config_preload),
             std::ref(_config_frontend),
             std::ref(_connections),
@@ -205,7 +205,7 @@ ss::future<> controller::start() {
             std::ref(_as));
       })
       .then([this] {
-          return _backend.start(
+          return _backend.start_single(
             std::ref(_tp_state),
             std::ref(_shard_table),
             std::ref(_partition_manager),