docs: add Neon AI Gateway endpoint guide

[mezotv]

Summary

Adds a Neon AI Gateway guide under the LibreChat YAML AI endpoints docs, plus the one sidebar entry in ai_endpoints/meta.json. Two files, insertion-only, no existing page or line is changed.

Neon AI Gateway is an OpenAI-compatible inference gateway, so the guide uses nothing but the endpoints.custom mechanism LibreChat already documents for OpenAI-compatible providers. It needs no application change, and it is modelled directly on the existing nearai.mdx page (#668).

The one thing that makes it different from the other custom-endpoint pages is that the gateway host is per Neon branch rather than per account, so the guide tells readers to add one custom endpoint per branch, each with its own name. It also covers the optional baseURL: "${NEON_GATEWAY_URL}" form, which works because endpoints.custom runs both apiKey and baseURL through extractEnvVariable.

fetch: true is used in the main example because the gateway implements GET /v1/models.

Two small choices worth flagging:

• Placeholder host. Every Neon branch host is different, so the example uses https://<your-neon-branch-host> rather than a real-looking hostname, matching the existing azure.mdx placeholder style. It sits inside code fences everywhere it appears so MDX does not read it as JSX.
• No icon: frontmatter key. I did not add a provider icon component. resolveIcon returns undefined for a missing key, so the sidebar entry degrades to text rather than breaking. Happy to add an icon in this PR if you would prefer the page to match its siblings.

Testing

Both CI jobs were reproduced locally on this branch:

• pnpm install --frozen-lockfile (lockfile unchanged)
• pnpm exec fumadocs-mdx
• pnpm lint:prettier — pass
• pnpm typecheck — pass
• pnpm lint — pass
• pnpm test — 184 tests in 21 files pass
• pnpm build — pass; the page is prerendered at /en/docs/configuration/librechat_yaml/ai_endpoints/neon and its llms.mdx route
• pnpm exec playwright test — 23 passed
• lychee with the exact arguments from .github/workflows/links.yml — 3 links, 3 OK
• Rendered the built page and confirmed the sidebar entry lands alphabetically between NEAR AI Cloud and NeurochainAI

Caveats

• No live end-to-end run. I have not driven a real LibreChat instance against a Neon branch. The configuration is verified against LibreChat's own source for the custom-endpoint path and against Neon's published documentation, not against a live chat session.
• Not a full-feature claim. The page documents chat completions and model listing, and deliberately claims nothing about embeddings, tool calling, structured output, or image generation. In particular, LibreChat's RAG API is a separate service with its own embeddings provider, which this endpoint does not cover; say the word if you would like an explicit sentence about that on the page.

Translations

English only. Per the README, .mdx is the source of truth and the locale files are generated, so I have left the translations to the pnpm translate workflow rather than hand-writing them.

[vercel]

@mezotv is attempting to deploy a commit to the LibreChat's projects Team on Vercel.

A member of the Team first needs to authorize it.