implement the use of multiprocessing for training on multiple envs

[sergiyayf]

• [x] Get first tests to on multiprocessing on a single node to work

[sergiyayf]

I run a small simulation with 2 environment for few steps to see if multiprocessing speeds up training, and it looked like it does. But then I run one with 10 environments overnight, and at the end the number of steps was the same if run on 1 env. PhysiCell simulation steps increased almost by 10.

ToDo:

• [x] Test speed up with the other 2 environments.
• [x] If it works with the other envs, but still doesn't with PhysiCell add ZMQ communication to a toy environment and see if it comes from there

[sergiyayf]

With GridEnv, using all of the cores for the job -> using SubprocVecEnv with 4 frames to stack -> double the speed. Probably, because you get speed up of running single env on multiple cores. and then for 10 ones, they have less resources per environment