[rllib] Move a3c implementation from examples/ to python/ray/rllib/ (#698)

* rllib v0

* fix imports

* lint

* comments

* update docs

* a3c wip

* a3c wip

* report stats

* update doc

* name is too long

* fix small bug

* propagate exception on error

* fetch metrics

* fix lint

doc/source/example-a3c.rst

@@ -25,7 +25,7 @@ You can run the code with
 
 .. code-block:: bash
 
-  python ray/examples/a3c/driver.py [num_workers]
+  python/ray/rllib/a3c/example.py --num-workers=N
 
 Reinforcement Learning
 ----------------------
@@ -153,6 +153,6 @@ workers, we can train the agent in around 25 minutes.
 
 You can visualize performance by running
 :code:`tensorboard --logdir [directory]` in a separate screen, where
-:code:`[directory]` is defaulted to :code:`./results/`. If you are running
+:code:`[directory]` is defaulted to :code:`/tmp/ray/a3c/`. If you are running
 multiple experiments, be sure to vary the directory to which Tensorflow saves
-its progress (found in :code:`driver.py`).
+its progress (found in :code:`a3c.py`).