Added support for stopping by max episodes instead of max steps.

[michalgregor]

• Added an additional argument to fit in core.py, which allows stopping by the number of episodes.
• Arguments nb_steps and nb_max_episode_steps are both optional now, but at least one must be specified, otherwise an exception is raised.
• Documentation of the relevant methods has been updated to reflect the new behaviour.
• TrainEpisodeLogger and TrainIntervalLogger have been updated.
• A minor fix to TrainIntervalLogger, which makes it print the statistics about any extant episodes even if the last logging interval ended prematurely.

[mirraaj]

@michalgregor , I will be looking into your implementation coming weekend. Thanks

[CameronBeebe]

This is still not implemented, any updates?