* fix cli args to config * fix train cli
* move s2t data preprocess into paddlespeech.dataset * avg model, compute wer, format rsl into paddlespeech.dataset * fix format rsl * fix avg ckpts