fix crash with return alternative on CUDA (OpenNMT#1733)

Author: minhthuc2502

src/decoding.cc

@@ -70,9 +70,14 @@ namespace ctranslate2 {
                                                   StorageView& ids) {
     if (!decoder.output_layer_is_updated())
       return;
+    ctranslate2::Device device = ids.device();
+    if (device != Device::CPU)
+      ids = ids.to(Device::CPU);
     auto* ids_data = ids.data<int32_t>();
     for (dim_t i = 0; i < ids.size(); ++i)
       ids_data[i] = decoder.to_original_word_id(ids_data[i]);
+    if (ids.device() != device)
+      ids = ids.to(device);
   }
 
   template <typename T>