arguments now don't have default values, because in continuation mode, we can't decide whether to change a parameter or not, because we don't know if a parameter is given or default
this is why default values has been moved to SequenceTaggerNetwork.__init__()
changed __get_state__ to save all training params into model dump
changed the way optional arguments (from argparse) change model's actual parameters in trainer.py
SequenceTaggerNetwork.__init__()__get_state__to save all training params into model dump