Add preprocessor/postprocessor normalizer files and train_config

ACT policy trained on combined third-train1+third-train2 (70 eps, 30k steps)

initial commit