Flux Pendulum DDPG example fails on GPU

OMG, this works like charm :see_no_evil:
I actually like the Flux model zoo implementation because everything is in one spot, this is much easier to grasp (for me at least). I will check if performance is better when adjusting for ACTION_BOUNDS in the actor network or when calling the environment and then make a pull request to have an updated version in the Model zoo.
Thanks a lot!