[BugFix] Force _get_batch_preds() to keep logits in decoder timestamps generator

[tango4j]

What does this PR do ?

The argument of _get_batch_preds in FrameBatchASRLogits class has not been updated when FrameBatchASR is updated.
Now, this PR force _get_batch_preds() to keep logits in decoder timestamps generators.

Collection: [Note which collection this PR will affect]
ASR

Changelog

• Added force_keep_logits for _get_batch_preds function in class FrameBatchASRLogits in decoder_timestamp_utils.py
• Changed class name FrameBatchASR_Logits to FrameBatchASRLogits

Usage

• You can potentially add a usage example below

python offline_diar_with_asr_infer.py \
    diarizer.manifest_filepath=<path to manifest file> \
    diarizer.out_dir='demo_asr_output' \
    diarizer.speaker_embeddings.model_path=<pretrained modelname or path to .nemo> \
    diarizer.asr.model_path=<pretrained modelname or path to .nemo> \
    diarizer.asr.parameters.asr_based_vad=True \
    diarizer.speaker_embeddings.parameters.save_embeddings=False

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

ASR contributors

Additional Information

This is for main, and there will be another PR for r1.18

[jbalam-nv]

LGTM