From homeostasis to resource sharing: Biologically and economically aligned multi-objective multi-agent gridworld-based AI safety benchmarks

Various smaller adjustments and improvements Adding Zenodo association Adding support for imitation learning / expert override with PPO multi-agent weight sharing setup. Adding configuration files for imitation learning. Also adding configs for 2-layout trials. Adding support for use_expln PPO and A2C policy argument to mitigate NaNs in SB3 tensors. Adding support for SB3 AdamW optimizer, needed to mitigate occasional PPO NaN's during imitation learning. Adding support for target_kl PPO config argument to mitigate NaNs in SB3 tensors. Adding support for early_detect_nans config argument to early detect NaNs in SB3 tensors. Adding support for soft_stop_training_on_nan_errors to handle NaNs in SB3 tensors smoothly. Adding retry_training_on_nans_for_n_times and skip_test_on_training_stop_on_nan_errors config settings. Adding new fields to the jsonl log: num_actual_train_episodes, training_run_was_terminated_early_due_to_nans, num_training_retries_used, test_checkpoint_filenames.