Fix hypothetical consumer deadlock

In the case when the user calls `AsyncClose` at the same instant as the broker
leading the partition returns an error.

Fixes bug #475.

consumer.go

@@ -409,14 +409,23 @@ func (child *partitionConsumer) responseFeeder() {
 			// so shut it down and force the user to choose what to do
 			Logger.Printf("consumer/%s/%d shutting down because %s\n", child.topic, child.partition, err)
 			child.sendError(err)
-			child.AsyncClose()
+			select {
+			case child.dying <- nil:
+			case <-child.trigger:
+			}
 		case ErrUnknownTopicOrPartition, ErrNotLeaderForPartition, ErrLeaderNotAvailable:
 			// these three are not fatal errors, but do require redispatching
-			child.dying <- err
+			select {
+			case child.dying <- err:
+			case <-child.trigger:
+			}
 		default:
 			// dunno, tell the user and try redispatching
 			child.sendError(err)
-			child.dying <- err
+			select {
+			case child.dying <- err:
+			case <-child.trigger:
+			}
 		}
 
 		child.broker.acks.Done()