Bump Automodel to Nemorl dedicate branch

## Title
Create a NeMo-RL dedicate branch and bump Automodel to [the branch](https://github.com/NVIDIA-NeMo/Automodel/tree/ruit/nemor-rl-submodule)

## Background / Motivation
- Align with upstream Automodel fixes/features to improve compatibility, performance, and stability.
- Target commit: `2d20e33a19d5e53a271b1403b507475e68ad14dc` (the “target commit”).
- Upstream repository: [Automodel Upstream]([https://github.com/NVIDIA-NeMo/Automodel]).

## Scope
- Upgrade Automodel to the target branch.
- Update lockfiles and dependency overrides in this repo as needed.
- Verify compatibility with existing training/inference pipelines (including vLLM integration, DTensor/MoE, etc.).

## Implementation Steps
1) Create a special branch for nemorl and point Automodel to the target commit
- `git submodule update --init path/to/automodel`
- In submodule: `git fetch origin && git checkout 2d20e33a19d5e53a271b1403b507475e68ad14dc`
- In main repo: `git add path/to/automodel && git commit -m "chore(automodel): bump to 2d20e33..."`
- Run `uv lock` to refresh lockfiles.

2) Dependency alignment and preinstallation

3) Configuration and environment

## Checklist
- [ ] Point to the target branch (submodule).
- [ ] `uv lock` and installation complete without warnings.
- [ ] Unit, integration, and performance tests pass.
- [ ] Docs and release notes updated (including rollback guidance).