fix general op reader problem, make fetch var changable during inference; add more arugments for resource

make general_model_config unified on client side and server side
make general_model_config accessible with Python API