feat: DPO support for global padding of seq_len to a multiple

[terrykong]

What does this PR do ?

• adds pad_to_multiple_of for DPO which is a requirement for sequence parallel (which is required for moe models w/ TP)
  • the argument pad_length_to_multiple_of will pad all minibatches to the same length if >0. If ==0, the behavior is the same as before.

Needed for:

• sequence parallel
• mamba

Rebase stack

• https://github.com/NVIDIA/NeMo-Aligner/pull/386
• https://github.com/NVIDIA/NeMo-Aligner/pull/387
• https://github.com/NVIDIA/NeMo-Aligner/pull/388
• https://github.com/NVIDIA/NeMo-Aligner/pull/380

Changelog

• Please update the CHANGELOG.md under next version with high level changes in this PR.

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this 

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? Make sure to also update the NeMo Framework User Guide which contains the tutorials

Checklist when contributing a new algorithm

• [ ] Does the trainer resume and restore model state all states?
• [ ] Does the trainer support all parallelism techniques(PP, TP, DP)?
• [ ] Does the trainer support max_steps=-1 and validation?
• [ ] Does the trainer only call APIs defined in alignable_interface.py?
• [ ] Does the trainer have proper logging?

Additional Information

• Related to # (issue)