Update Gemma3 multimodal SFT Jupyter notebook

[SurbhiJainUSC]

Description

This PR updates the Gemma 3 multimodal Supervised Fine-Tuning (SFT) Jupyter notebook to align with the latest MaxText training utilities and the native SFT pipeline. The primary goal is to provide a streamlined, end-to-end example for users to convert checkpoint, and fine-tune Gemma 3 multimodal model.

Key Changes:

• Notebook Update: The multimodal_gemma3_demo.ipynb has been updated to use maxtext.trainers.post_train.sft.train_sft_native.
• Workflow Enhancements: Updated the Checkpoint Conversion section to demonstrate converting Orbax checkpoints from Hugging Face for Gemma 3 models.
• Trainer Alignment: Minor updates to train_sft_native.py to better align with train.py.
• Dependency Management: Updated instructions for installing post-training dependencies within the notebook environment.

Tests

• Notebook Execution: Verified the end-to-end flow in VsCode on TPU v5p-8.
• Checkpoint Conversion: Confirmed successful conversion of Gemma3-4B weights from Hugging Face.
• SFT Run: Successfully executed a 10-step SFT run on the ChartQA dataset using the native trainer.

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[YixuanWang-99]

LGTM