This recipe implements Streaming Zipformer-Transducer model.

See https://k2-fsa.github.io/icefall/recipes/Streaming-ASR/librispeech/zipformer_transducer.html for detailed tutorials.

[./emformer.py](./emformer.py) and [./train.py](./train.py)
are basically the same as
[./emformer2.py](./emformer2.py) and [./train2.py](./train2.py).
The only purpose of [./emformer2.py](./emformer2.py) and [./train2.py](./train2.py)
is for exporting to [sherpa-ncnn](https://github.com/k2-fsa/sherpa-ncnn).