[Training] Allow the training task to accept pre-trained RL-agents as competitors

[NikkelM]

Currently, the training task will not accept pre-trained RL-agents as competitors, as they need an argument (their modelfile), which would be rejected by the environment_config.py when parsing the agents.