Defer vLLM and SGLang GPU pages to their engine cookbooks

[vincentzed]

Rewrite the vLLM and SGLang GPU-inference pages to lead with the engine cookbook and keep only a minimal quickstart, so the model- and hardware-specific details live in one continuously maintained place instead of drifting here.

What changed

• vLLM (deployment/gpu-inference/vllm.mdx) and SGLang (deployment/gpu-inference/sglang.mdx): each page now opens with a Card to its cookbook (vLLM, SGLang), a scannable "What's in the cookbook" list, a generic install/serve/query quickstart, and a sampling note. The full model matrix, per-model serve flags (tool calling, reasoning, vision), recommended-sampling tables, Blackwell tuning, and benchmarks are deferred to the cookbooks.
• Nav (docs.json): reorder the GPU Inference group so SGLang sits above vLLM.

Notes

• SGLang links to the canonical docs.sglang.ai cookbook (not the mintlify.app preview).
• Both pages keep their existing URLs and stay in the nav, so incoming cross-links and the link-snapshot ledger are unaffected.

Validation

• npm run snapshot:check → 170 active URLs verified; pre-commit snapshot hook clean.
• No self-closing MDX tags; copy contains zero em/en-dashes.
• Rendered both pages locally on mintlify dev → 200.

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
liquid-docs	🟢 Ready	View Preview	Jun 26, 2026, 10:14 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.

[alay2shah]

lgtm