Introduce new optional "relay" backend docker image that can be self-hosted to relay AI requests

[Logende]

AI backend / relay

Create a small self-hosted backend relay for MC that forwards requests to a user-configured LLM endpoint with a user-configured API key. This solves browser CORS issues for providers that cannot be called directly from the frontend, while keeping provider keys out of the browser.

Scope

• Minimal Flask app.py + simple Dockerfile.
• Config via environment variables only.
• Intended for self-hosting by individuals, teams, or companies.

Required config

• UPSTREAM_BASE_URL
• UPSTREAM_API_KEY
• MC_RELAY_USERNAME
• MC_RELAY_PASSWORD
• Optional: ALLOWED_ORIGIN, timeout, extra upstream headers.

Required features

• HTTP Basic Auth for MC -> relay access.
• CORS support for the configured MC origin.
• Proxy OpenAI-compatible endpoints, at minimum:
  • GET /health
  • GET /v1/models
  • POST /v1/chat/completions
• Pass upstream responses through with minimal transformation.
• Support streaming if possible.

Security requirements

• No provider key in frontend/browser.
• No secrets in logs.
• Deploy behind HTTPS.
• Restrict CORS to configured origin, not *.

Goal

Add support for custom/self-hosted and CORS-blocked LLM providers via a simple relay that can live inside the MetaConfigurator repo and be easy to run with Docker.

Also sufficiently document how to set up and use this "relay".