Add new rllib agent

[oliverwang15]

• The new agent adapts to ray >= 1.13.0
• Changes in DRL_prediction function. The function returns a rllib_agent which can output actions according to the saved model and input observations so that we can customize our backtest process with our environment now.
• You may find the demo code on using the agent here.

[zhumingpassional]

thanks for your effort.

if the new code is better, can we replace the old version?

[oliverwang15]

Hi! I have replaced the old version with the new one. In case this action make conflicts, I kept the old version below and renamed it with DRLAgent_old. We may delete it if no one raises any issues after several months. Please check it out again. Thank you!

[zhumingpassional]

ok