Add RL GRPO gemma4-26b tutorial and fix doc conversion links

[dasoto]

Description

This PR adds a Reinforcement Learning (GRPO) tutorial and launch script for gemma4-26b (26B-A4B MoE) on Ironwood (Cloud TPU v7x), following the same format as the existing gemma4-e4b recipe. It also fixes several broken documentation links in the existing RL tutorials and registers the per-model Gemma RL tutorials in the docs table of contents.

What's added

• docs/tutorials/posttraining/rl_gemma4_26b.md — step-by-step tutorial to run gemma4-26b GRPO on the OpenMathInstruct-2 dataset on a tpu7x-4x4x4 (64-chip) slice. Mirrors the gemma4-e4b tutorial structure: prerequisites, HF auth, checkpoint conversion, chat-template setup, workload submission, monitoring, and export back to HF.
• src/maxtext/trainers/post_train/rl/scripts/run_gemma4_26b_rl.sh — XPK/Pathways launch script for the recipe. Mirrors run_gemma4_e4b_rl.sh and applies the 26B-specific configuration (model gemma4-26b, base tokenizer google/gemma-4-26B-A4B, MoE rollout expert-parallelism, tpu7x-4x4x4 slice, pinned Pathways images, etc.).

Docs housekeeping (fixes to existing RL tutorials)

• Registered rl_gemma4_26b.md and the previously-orphaned rl_gemma4_e4b.md in post_training_index.md (RL list + toctree).
• Fixed a broken relative link to the build guide (../../build_maxtext.md → ../build_maxtext.md).
• Fixed broken MyST cross-references: the bare [...](hf-to-maxtext) / [...](maxtext-to-hf) links did not resolve. Replaced them with the working file+anchor form (../../guides/checkpointing_solutions/convert_checkpoint.md#hugging-face-to-maxtext and #maxtext-to-hugging-face), matching the pattern already used in dpo.md / lora.md / rl_on_multi_host.md.
• These link fixes were applied across rl.md, rl_gemma4_e4b.md, rl_gptoss_20b.md, and rl_qwen3_30b.md (scoped to the RL tutorials only).

Tests

Please describe how you tested this change, and include any instructions and/or
commands to reproduce.

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.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.