Tolerate race condition when starting management db cache process

This prevents the below harmless crash when multiple parallel API
requests arrive soon after starting the node.

```
exception error: no match of right hand side value
                 {error,{already_started,<0.1593.0>}}
  in function  rabbit_mgmt_db_cache:fetch/4 (rabbit_mgmt_db_cache.erl, line 68)
  in call from rabbit_mgmt_db:submit_cached/4 (rabbit_mgmt_db.erl, line 756)
  in call from rabbit_mgmt_util:augment/2 (rabbit_mgmt_util.erl, line 412)
  in call from rabbit_mgmt_util:run_augmentation/2 (rabbit_mgmt_util.erl, line 389)
  in call from rabbit_mgmt_util:augment_resources0/6 (rabbit_mgmt_util.erl, line 378)
  in call from rabbit_mgmt_util:with_valid_pagination/3 (rabbit_mgmt_util.erl, line 302)
  in call from rabbit_mgmt_wm_queues:to_json/2 (rabbit_mgmt_wm_queues.erl, line 44)
  in call from cowboy_rest:call/3 (src/cowboy_rest.erl, line 1583)
```

deps/rabbitmq_management/src/rabbit_mgmt_db_cache.erl

@@ -65,9 +65,14 @@ fetch(Key, FetchFun, FunArgs, Timeout) ->
     ProcName = process_name(Key),
     Pid = case whereis(ProcName) of
             undefined ->
-                {ok, P} = supervisor:start_child(rabbit_mgmt_db_cache_sup,
-                                                 ?CHILD(Key)),
-                P;
+                case supervisor:start_child(rabbit_mgmt_db_cache_sup,
+                                            ?CHILD(Key)) of
+                    {ok, P} ->
+                        P;
+                    {error, {already_started, P}} ->
+                        %% A parallel request started the cache meanwhile
+                        P
+                end;
             P -> P
           end,
     gen_server:call(Pid, {fetch, FetchFun, FunArgs}, Timeout).