docs: add API references for 28 newly launched models

[liyangbing]

Summary

Adds Model API Reference pages for 28 models that are live in the public catalog (/api/models) but were missing from the docs.

• LLM (10): gpt-6-astra, claude-fable-5.1, gemini-3.8-flash, deepseek-v4.1-flash, deepseek-v4-flash-vision-exp, qwen3.8-27b, hy4-preview, jev-1.13 (new vendor TypeSafe), solar-pro4, glm-5.3
• Image (13): midjourney-v8.1 (new vendor), gpt-image-2.5-flare / sunburst (+ edit), Topaz adjust / denoise / restore / sharpen / upscale (creative, generative, precision, transparent)
• Video (5): wan 3.0 video / prime, minimax h3-max text / image / reference-to-video

Sidebar and category overview pages now include these models. The change only adds content. Existing model pages, Platform API pages, and hand-normalized overview wording are unchanged.

Generator fixes (scripts/generate-llm-model-api-reference.mjs)

• Sidebar vendor groups use vendorLabel(), so registry spellings like Bytedance render as ByteDance.
• OpenAI models now report /v1/responses in SEO metadata, matching the page body. The metadata previously said /v1/chat/completions.
• Gemini Interactions models are skipped in LLM Models because they are documented only under Official Native API (see a61180f). This also removes a sidebar link to the non-existent llm-models/google/gemini-omni-flash-preview page.

Not included

• *-exp0622, seedance-2.x-official, and gpt-image-2.5-*-official entries. They are enabled in the registry but absent from the public catalog.
• 37 video models that the registry has reclassified from image to video. They already have image-generation pages, so moving them belongs in a separate change.

Verification

• npm run validate passes (OpenAPI lint, public API, links, SEO, AI-readable, doc content, worker, build, built surface).
• npm run test:generator has 2 failures that also occur on main without this change.