fix(docs): refresh DeepSeek V4 verification pages

[yaoyu-33]

Summary

• regenerate the model verification catalog after PR feat(recipes): modernize DeepSeek V4 library configs #5554 updated the DeepSeek V4 verification card
• refresh the Sphinx and Fern DeepSeek V4 model pages
• restore the generated-output consistency gate failing on main

Closes the failure in https://github.com/NVIDIA-NeMo/Megatron-Bridge/actions/runs/33416245103/job/99569478293.

Testing

• uv run --no-project --with pyyaml python scripts/docs/generate_model_verification_catalog.py --check
• git diff --check
• uv run --no-project --with pre-commit pre-commit run --all-files

The focused pytest entry point cannot initialize on macOS because the shared MCore conftest imports Triton, which has no supported macOS wheel. The exact failing generator assertion was reproduced before regeneration and passes afterward.

[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.

[claude]

Light review - LGTM

Faithful regeneration of the generated DeepSeek V4 verification catalog and doc pages from source card.yaml, restoring the generated-output consistency gate.

Checks: metrics, dates, statuses, precision (bf16 to fp8_mx for pretrain/resume), commands, and enabled_features in catalog-v1.json and both doc pages match the source card.yaml. The two rendered docs (deepseek-v4.md and deepseek-v4.mdx) are byte-identical, as the generator intends. All three newly-referenced recipes (deepseek_v4_flash_pretrain_64gpu_gb200_fp8mx_library_config, deepseek_v4_flash_sft_openmath_thinking_packed_gb200_config, deepseek_v4_flash_peft_openmath_thinking_packed_gb200_config) resolve to real modules under src/megatron/bridge/recipes/deepseek/. PEFT item correctly flips unverified to verified with populated metrics and a runnable command; stale placeholders are cleared consistently across all three files.

No critical bugs, typos, or inaccuracies found. Docs-only, generated-artifact PR.

Suggested test cases

• No perf tests impacted. This PR touches only catalog-v1.json and the two generated DeepSeek V4 doc pages; no scripts/performance/configs/ files or recipe configs are changed.
• Recommended CI gate: scripts/docs/generate_model_verification_catalog.py --check (the generated-output consistency check this PR restores).