feat: Add script to re-initialize near-zero HF embeddings

[ashors1]

What does this PR do ?

For certain models (e.g. nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16), the HF checkpoint has embedding weights which are zero or near-zero. These weights can cause inf/nan grad norms during post-training. This PR adds a script to reinitialize such embeddings. It also adds an inf grad norm check with an informative message.

Issues

List issues that this PR closes (syntax):

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 run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.