Update for the latest library enviromnet

[Lukious]

Hello, first of all, thank you for providing this great project.

As I was referring to your repository to write the code, I found that it has not been updated for quite a long time. I have modified some parts of the code to make it suitable for the latest gym library environment (changes include Acer, Actor_critic, DQN, PPO, PPO-LSTM, REINFORCE, and VTRACE algorithms).

Some corrections have been made for the tuple-type values returned from the updated Open AI Gym's dependency, and for the conversion speed due to data type transformations.
(e.g. UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor.)

Most of the algorithms' speed has been improved by more than 10%, and we have made minimal changes to maximize efficiency, keeping in line with the spirit of minimalRL.

Thank you again for this great project!

p.s. Add some detailed patch note

Algorithm	State	Etc.
A2C	Not-updated (Pipe-line error)	-
A3C	Pending Update	To be update
Acer	Update done (performance improved)	-
Actor-critic	Update done (performance semi-improved)	-
DDPG	Not-updated (Pendulum-v1 is not working)	-
DQN	Update done (performance improved)	-
PPO	Update done (performance improved most)	Performance improved more than 20%
PPO-continuous	Not-updated (Pendulum-v1 is not working)	-
PPO-LSTM	Update done (performance improved)	-
REINFORCE	Update done	-
SAC	Not-updated (Pendulum-v1 is not working)	-
VTRACE	Update done (performance improved)	-

#59 is solved now.

[Lukious]

I checked the repo master's new release.

#59 Has been solved the same way.

Thanks for the update.