Refactor the optimizer override function so that users can swap in their own

[jstjohn]

What does this PR do ?

Makes the private optimizer.py:_build_config_overrides into a builder object that a user can swap in their config. This allows new models (eg evo2) to have their own custom override function without adding booleans/modifying the core function in mbridge optimizer.py.

Changelog

• Adds OptimizerConfigOverrideProvider to the ConfigContainer that replaces the _build_config_overrides function.
• Adds missing handling of decoupled lr to the config override builder
• Adds note about how we can simplify further once [main] feat(moe): Support apply wd to qk layernorm for Qwen3-Next (4/4) NVIDIA/Megatron-LM#2753 is pulled into dev.
• Builds off of enable qwen wd #1935
• Depended on by Refactor the way that we do weight decay skipping for hyena to follow ToT mbridge. NVIDIA/bionemo-framework#1429

GitHub Actions CI

See the CI sectionin the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.

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?

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

Additional Information

• Related to # (issue)

[FDecaYed]

Agree on making the config override not hard coded and each model need to change something create and pass their own. we probably also want to handle qwen the same way instead of hardcode if block in default