AttributeError: PyProcessDmLab object has no attribute b'initial'

I am trying to run the single-level experiment as instructed in Readme:
```
python experiment.py --num_actors=48 --batch_size=32
```

But right after all the 48 actors are created the script crashes with the following stacktrace:

```
File "/home/apetrenk/miniconda3/envs/doom-impala/lib/python3.7/site-packages/tensorflow/python/ops/script_ops.py", line 209, in __call__
ret = func(*args)

File "/home/apetrenk/all/projects/doom/impala/scalable_agent/py_process.py", line 87, in py_call
raise result

AttributeError: 'PyProcessDmLab' object has no attribute 'b'initial''


[[node flow_environment_initial_3/initial (defined at /home/apetrenk/all/projects/doom/impala/scalable_agent/py_process.py:97) ]]
```

Looks like for some reason the byte-literal is used instead of a string for the method name.

Reproduce the failure with the README command `python experiment.py --num_actors=48 --batch_size=32`. Start in `scalable_agent/py_process.py` around `py_call` and the `flow_environment_initial_3/initial` call, then trace how the method name reaches `PyProcessDmLab`. Done means the single-level experiment starts without the reported AttributeError.

Written by the indexing model from the issue text.