Adapted gym env to only del a visualizer if the environment name is in the dictionary, as this is not the case during intialization of the rl env.

Inference uses the same configs as the training and instantiates the model load function from the configs.

Fixed a bug with the training configs.