docs(models): expand GLM-5.3-Flash coverage

[HuiyingLi]

Summary

• announce Ox Alpha / GLM-5.3-Flash in the README and add its supported-model recipe entry
• expand the model coverage page with the hybrid KDA / KPool-DSA architecture, 288-expert topology, supported training path, dependencies, and current limitations
• document the validated MedPix packed-2K EP72/CP2 recipe, 4K Hugging Face logit parity, packed CP1/CP8 training parity, and SDPA versus cuDNN performance with W&B evidence
• clarify that EP72 is a validated topology rather than a model requirement, while full-model single-GPU loading, TP, PP, and video training remain unsupported

Context

This is the docs follow-up to the merged implementation in #3699. That PR added the nightly navigation entry and a minimal coverage page; this PR completes the standalone user documentation without changing model code or recipes.

Validation

• make -C docs/fern docs-check — 477 MDX files validated, 0 errors
• .venv/bin/python -m pytest -q tests/unit_tests/tools/test_sync_model_coverage_tables.py — 22 passed
• git diff --check

Signed-off commit: a702ab2d665e5c4c2c8d46f5da86955449f006a2

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

🌿 Preview your docs: https://nvidia-preview-preview-a702ab2d665e.docs.buildwithfern.com/nemo/automodel

[HuiyingLi]

/ok to test a702ab2