Add MoE load balancing loss to distillation

[JamesDeng42]

Description

This PR introduces support for Mixture of Experts (MoE) load balancing loss during the distillation workflow.

Key Changes

1. NNX Intermediate Extraction (maxtext_utils.py & qwen3.py):
   • Replaced legacy Linen self.sow(...) calls with native nnx.Intermediate(load_balance_loss) inside
2. Distillation Strategy Updates (distillation_utils.py & train_distill.py):
   • Upgraded DistillationForwardOutput to carry the collected moe_lb_loss.
   • Updated CombinedDistillationStrategy to actively add the moe_lb_loss to the total_loss so the optimizer
     minimizes it.
   • Surfaced "distill/moe_lb_loss" to the metrics dictionary for TensorBoard logging and visibility.
3. Model Mutability (models.py):
   • Automatically appended "intermediates" to the mutable_collections list during the Transformer's forward pass
     whenever load_balance_loss_weight > 0.0 to ensure NNX variables can successfully write to the state.

Tests

Added "distill/moe_lb_loss" to the expected metrics keys in the test suite to prevent regressions in train_distill_test.py.

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

❌ Patch coverage is 22.22222% with 14 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...ners/post_train/distillation/distillation_utils.py	37.50%	3 Missing and 2 partials ⚠️
.../trainers/post_train/distillation/train_distill.py	16.66%	4 Missing and 1 partial ⚠️
src/maxtext/models/models.py	0.00%	1 Missing and 1 partial ⚠️
src/maxtext/models/qwen3.py	0.00%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!

[vlad-karp]

please address the comment

[gagika]

thanks