Support using the branch from the gigaspeech dataset for decoding

[csukuangfj]

WER comparison

	test-clean	test-other	comment
basline	2.78	7.36	--iter468000 --avg 16, greedy search
this PR	3.36	8.20	--iter 468000 --avg 16, greedy search

It performs better if we use the branch from the librispeech for decoding. However, it is not clear how it performs on other dataset if we use the branch from the gigaspeech for decoding.

The decoding results are listed below

	test-clean	test-other
baseline	errs-test-clean-greedy_search-iter-468000-avg-16-context-2-max-sym-per-frame-1-use-averaged-model.txt	errs-test-other-greedy_search-iter-468000-avg-16-context-2-max-sym-per-frame-1-use-averaged-model.txt
this PR	errs-test-clean-greedy_search-iter-468000-avg-16-use-giga-branch-context-2-max-sym-per-frame-1-use-averaged-model.txt	errs-test-other-greedy_search-iter-468000-avg-16-use-giga-branch-context-2-max-sym-per-frame-1-use-averaged-model.txt

[csukuangfj]

The pretrained models are uploaded to https://huggingface.co/csukuangfj/icefall-asr-librispeech-lstm-transducer-stateless2-2022-09-03/tree/main/exp/giga

[danpovey]

My wife was saying then when she says things like "google" on the librispeech output, it was never getting it right. We'll see whether the gigaspeech one is better.