Parl

Frameworks
- Support tensorboard tool.
- Add `save` and `restore` APIs in `parl.Agent`.
- Add exception traceback in remote module.
- Disentangle basic classes(e,g., parl.Model) and the computation framework.

Examples
- Refine benchmark performance of A2C example.
- Simplify QuickStart example.

Papers
- Collect some papers relative to model-based reinforcemnt learning topic.

Documentation
- Add Chinese version of README in homepage.

Framework
- Support for distributed training. Add parallelization module `parl.remote`.
- Functional APIs to dump and load parameters in numpy arrays. Add `get_params` and `set_params` to support getting parameters from `parl.Model`, `parl.Algorithm` and `parl.Agent`.
- Add IMPALA and A3C algorithms to `parl.algorithms`.

Examples
- IMPALA
- A2C
- GA3C

Framework
- Support `Model`, `Algorithm` and `Agent` abstractions.
- Support wrappers for fluid.layers, which can easily share parameters between layers.
- Support `sync_params_to` API in `Model` to synchronize parameters between model and target model directly.

Examples
- QuickStart
- DQN
- DDPG
- PPO
- Winning solution of NeurIPS2018-AI-for-Prosthetics-Challenge