Adding fix to absolute embedding

[tbartley94]

What does this PR do ?

Fixes ASR transformer embedding behavior to allow adaption to strings larger than max_sequence_length.

Before, transformer module assumed that Absolute Positional Embeddings resized for cases where seq_length > max_sequence_length. This actually was not implemented and had no safeguards, leading to indexing errors in edge cases for Canary decoding.

Now automatically resizes and adds better logging to alert user.

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

Changelog

• Refactored transformer_modules.FixedPositionalEncoding so the absolute positional information is a standalone method
• Added check to FixedPositionalEncoding to resize when position_ids tokens are longer than max_sequence_length
• Added flags to alert user when seq_length > max_sequence_length iin TransformerEmbedding module

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR. To re-run CI remove and add the label again. To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

• [y ] Make sure you read and followed Contributor guidelines
• [ n] Did you write any new necessary tests?
• [ n] Did you add or update any necessary documentation?
• [ n] 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:

• [ y] New Feature
• [y ] Bugfix
• [ ] Documentation