docs(nnx): add native LoRA Gemma4 and Qwen3 tutorial notebooks

[RexBearIU]

Description

Adds native Gemma4 and Qwen3 (0.6b) LoRA and QLoRA tutorial notebooks for Flax NNX.

These tutorials cover both:

1. Pathway A: SFT Fine-Tuning with native LoRA.
2. Pathway B: Pre-training with QLoRA, including checkpoint saving and restoration.

IMPORTANT NOTE: This pull request is blocked by PR #4346 (feat(nnx): support native Flax NNX PEFT/LoRA training loop). It targets main directly to keep its commit history clean, but should only be merged after PR #4346 has been successfully integrated into main.

BUGS: #4400

Tests

Both tutorial notebooks have been extensively verified and executed cell-by-cell under standard TPU VM environments using papermill with zero errors:

1. Qwen3 Tutorial Notebook (qwen3_native_lora_demo.ipynb)

   • Compiles, runs, saves checkpoints, and restores checkpoints for pre-training.
   • Validated with 100% cell execution success.

2. Gemma4 Tutorial Notebook (gemma4_native_lora_demo.ipynb)

   • Configured with scan_layers=False to handle the specific multi-query KV sharing architectural constraints of Gemma4-e2b.
   • Compiles, runs, saves checkpoints, and restores checkpoints.
   • Validated with 100% cell execution success.

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!