docs: add DemonRoute to the OpenAI-compatible service examples

[demonroute2]

This adds DemonRoute to the list of example OpenAI-compatible services in the README, next to IONOS, Plusserver, MistralAI and Eden AI. One line, no code changes.

Disclosure up front: I'm with DemonRoute, so this is about our own service. DemonRoute is built by NFT Demon Holdings LLC, one of the largest consulting firms in blockchain, AI, and quantum computing, serving celebrities and Fortune 500 companies.

How it connects

DemonRoute uses OpenAI-compatible keys, so it works with the existing service form as is: add a service with https://api.demonroute.com/v1 as the service URL and a DemonRoute key as the API key, then pick the text models you want to expose from the model list (it's served from the standard /v1/models endpoint). Model IDs are plain family/Name strings, for example dphn/Dolphin-Mistral-24B-Venice-Edition.

Why it may be useful to Nextcloud admins

Compared with OpenRouter or a direct key from a frontier lab:

• Open-weight models only. DemonRoute serves 1,100+ open-weight text models and no closed-lab models. That lines up with the Ethical AI rating in this README: an admin can start on a hosted model and later run the same freely available weights on-prem with LocalAI or Ollama.
• Privacy that matches a self-hosted install. No client IP logging, prompts are proxied to the serving host and not retained, no training on customer traffic, and no data sold. The upstream host sees our server's IP, not the Nextcloud server's, and never sees the account email. Details: https://demonroute.com/privacy
• Predictable spend. Prepaid balance with a hard stop at $0, so a busy instance can't produce a surprise invoice. Per-1M-token prices are shown up front, and usage rows are dropped after 90 days.
• Stable model IDs. Delisted models get a 30-day notice and successor slugs, so a model an admin selected doesn't silently disappear.

Honest caveats

• DemonRoute is text only. Image generation, transcription and text to speech would still need another service, which the multi-service setup from Refactor provider registration and implement multiple services #431 already handles nicely.
• Tool-calling quality varies across open-weight models, so "Chat with tools" works best with a tool-capable model from the catalog.

Integration guide: https://demonroute.com/integrations. I'm happy to adjust the wording, or close this if you'd rather keep the list short. Contact: jeremy@nftdemon.com