fix(go-grpc-server): always close resultChan

By not closing the channel, if a server not implementing PredictStream
receives a client call would hang indefinetly as would wait for
resultChan to be consumed.

If the prediction stream returns we close the channel now and we wait
for the goroutine to finish.

Signed-off-by: Ettore Di Giacinto <mudler@localai.io>

pkg/grpc/server.go

@@ -144,6 +144,8 @@ func (s *server) PredictStream(in *pb.PredictOptions, stream pb.Backend_PredictS
 	}()
 
 	err := s.llm.PredictStream(in, resultChan)
+	// close the channel, so if resultChan is not closed by the LLM (maybe because does not implement PredictStream), the client will not hang
+	close(resultChan)
 	<-done
 
 	return err