[s2t]  streaming conformer u2 and u2pp jit export

* update mix frontend, test=tts

[ASR] support u2pp based cli and server, optimiz code of u2pp decode (reversed_weight parameter)

[Server]Deploy text model in server

* BytesIO类型时，要保证切到初始位置，这样多次读取才能够正常。比如__call__函数。
__call__函数的参数audio_file为BytesIO类型时执行到self.preprocess(model, audio_file)会报错，需要判断audio_file为BytesIO类型时执行audio_file.seek(0)。

[ASR] Chang memory allocator strategy to fix gpu training hang

Text example config files

[CLI]Deploy fast text model for cli

[doc]add Speaker Diarization in readme