Created by: xinghai-sun

1. Move functions in model.py into layer.py
2. Add a DeepSpeech2Model class in model.py, which has train() and infer_batch() methods.
3. Simplify train.py, evaluate.py, infer.py and tune.py。