Support RNNLM rescoring for streaming models

[tempops]

Hello, I was looking at implementing RNNLM rescoring and saw that there was a RNNLM option for offline ASR, as stated here Add RNN LM rescore for offline ASR with modified_beam_search #125.

Is there an option to pass a language model option for streaming recognition? I do see that there is a --lm option for the OnlineRecognizer class in online_recognizer.py.

It will be great to get guidance and clarity on this. Thank you!