RNN-T constants

[mwawrzos]

1. Changes in mlperf_logging/mllog/constants.py reflect recent rules in https://github.com/mlcommons/training_policies/blob/47e08397ac6dd46a940816c65bd291a9d218383f/training_rules.adoc#91-hyperparameters
2. mlperf_logging/compliance_checker/1.0.0/closed_rnnt.yaml had to be updated after changes in 1.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[mwawrzos]

recheck

[mwawrzos]

@xyhuang I made the changes you proposed at the last logging meeting:

1. I have rewritten the checks to validate based on metadata here: aa44366
   An example of reporting such entry is here: https://github.com/mlcommons/training/blob/66d6c2dc784ff5b15641807a65036ffedd68af83/rnn_speech_recognition/pytorch/rnnt/model.py#L131-L132
2. I also worked with Michał to add Unet3D checks here in the v1.0.0 directory.
3. I moved the RNN-T constants update to RNN-T constants #84 to make this PR independent from RNN-T hyperparamters training_policies#425

Is it possible to merge this PR?

[mwawrzos]

@xyhuang , hyperparameter names are accepted in mlcommons/training_policies#425
Can we merge this PR now?