blaze convert dense model failed because the xdl version upgrade to 1.2

The `Nodetype` definition in `xdl(version=1.2) graph_def.proto` is
```
message NodeDef {
string name = 1;
string op = 2;
repeated string input = 3;
DeviceDef device = 4;
map attr = 5;
}
```

but in xdl(version=1.1) is
```
message NodeDef {
string name = 1;
string op = 2;
repeated string input = 3;
repeated DataType output_type = 4;
DeviceDef device = 5;
map attr = 6;
}
```

and in the same `graph_def.proto` of blaze is
```
message NodeDef {
string name = 1;
string op = 2;
repeated string input = 3;
DeviceDef device = 4;
map attr = 5;
}
```

So, there is a error in blaze conveting dense model.
```
[libprotobuf ERROR /home/work/open-code/x-deeplearning/blaze/thirdparty/protobuf/protobuf-3.6.0/src/google/protobuf/text_format.cc:307] Error parsing text-format xdl.proto.GraphDef: 4:14: Message type "xdl.proto.NodeDef" has no field named "output_type".
```

Start by comparing the xdl 1.2 and 1.1 NodeDef definitions in graph_def.proto with Blaze's definition. Reproduce the dense-model conversion and inspect the protobuf text-format error about the missing output_type field; done means the conversion no longer fails because of this schema mismatch.

Written by the indexing model from the issue text.