Rockrl

Added
- Added `ema_ratio` parameter to `rockrl.tensorflow.ppo` to control the exponential moving average ratio of the old and new networks

Changed
- Changed `rockrl.tensorflow.ppo` `act` function to use training mode while training and evaluation mode while evaluating

Added
- Added `close()` method to `rockrl.tensorflow.ppo` to close `tensorBoardLogger` after training

Changed
- Updated `rockrl.tensorflow.ppo` to use threading while logging, to save time

Changed
- Changed `rockrl.utils.VectorizedEnv` to return info dictionaries in list instead of tuples

Changed
- Removed `tensorflow_probability` dependency