Add stochastic depth to conformer and share results on LibriSpeech 960h

[pyf98]

Hi, in this PR, I added stochastic depth to conformer_encoder so that layer dropout can be used. I am also sharing two configs (with and without layer dropout) and their results for LibriSpeech 960h. As mentioned previously, the results are better than the current ones.

[pyf98]

Full results are coming soon. (Some decoding jobs are still running)

[codecov]

Codecov Report

Merging #4142 (dabbc3c) into master (6f42960) will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #4142      +/-   ##
==========================================
+ Coverage   80.43%   80.46%   +0.02%     
==========================================
  Files         442      442              
  Lines       38557    38562       +5     
==========================================
+ Hits        31015    31027      +12     
+ Misses       7542     7535       -7     

Flag	Coverage Δ
test_integration_espnet1	67.13% <ø> (ø)
test_integration_espnet2	51.13% <20.00%> (-0.01%)	⬇️
test_python	66.53% <100.00%> (+0.02%)	⬆️
test_utils	24.45% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
espnet2/asr/encoder/conformer_encoder.py	96.55% <100.00%> (+0.12%)	⬆️
espnet2/asr/transducer/beam_search_transducer.py	98.74% <0.00%> (+0.94%)	⬆️
...et/nets/pytorch_backend/conformer/encoder_layer.py	93.10% <0.00%> (+4.59%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 6f42960...dabbc3c. Read the comment docs.

[csukuangfj]

How many parameters does the model have?

[pyf98]

I have updated the decoding results and also uploaded models to huggingface. Please review the changes.