fix(metrics): fix race condition when calling Broker.Open() twice

On failure, `Broker.Open()` can be called several times while a producer
is running. In IBM#2409, it was assumed that AsyncProduce can only be
called with an open broker, however, it should be read that a user
should call it after opening the broker. The broker could be
disconnected and in progress of being reconnected. This is not hard to
fix as we already have a lock protecting the creation of the registry:
just don't create a new metric registry when attempting to reopen the
broker.

Fix IBM#2320 (again)

broker.go

@@ -175,7 +175,9 @@ func (b *Broker) Open(conf *Config) error {
 
 	b.lock.Lock()
 
-	b.metricRegistry = newCleanupRegistry(conf.MetricRegistry)
+	if b.metricRegistry == nil {
+		b.metricRegistry = newCleanupRegistry(conf.MetricRegistry)
+	}
 
 	go withRecover(func() {
 		defer func() {