Adding PQN Support

[LabChameleon]

PQN is becoming a popular off-policy RL algorithm that appears to show different behaviour concerning its hyperparameters compared to DQN. Additionally, it allows much faster training and does not require a replay buffer. This makes it interesting to be studied regarding HPO in addition to the currently implemented algorithms.

[TheEimer]

I pushed my current status for PQN on a separate PQN branch. It's running, but fully untested. I'm currently running experiments with it and in the first versions it seems to be similar to DQN on LunarLander. Probably testing and a thorough checkup is good.