feat(fastapi): add MiniMax LLM provider

[octo-patch]

Summary

Adds MiniMax as a new LLM provider for the server/fastapi classic STT → LLM → TTS voice pipeline.

MiniMax exposes an OpenAI-compatible chat-completions API, so the provider reuses Pipecat's OpenAILLMService pointed at the MiniMax base URL (https://api.minimax.io/v1). This is the same pattern Pipecat itself uses to wire other OpenAI-compatible vendors (Grok, Groq, DeepSeek), so it fits the existing provider architecture cleanly and needs no new dependencies.

Changes

• Add models/llm/minimax.py — thin provider module, default model MiniMax-M3
• Register "minimax" in the LLM provider catalog in models/providers.py (requires MINIMAX_API_KEY); it now shows up in the /providers endpoint and is validated at startup like the other providers
• Document MINIMAX_API_KEY (and the optional MINIMAX_BASE_URL override for the mainland-China endpoint) in env.example
• Update the FastAPI README.md (provider list, required keys, module list, example stack)

Usage

CLASSIC_LLM_PROVIDER=minimax
MINIMAX_API_KEY=your_minimax_api_key

The default model is MiniMax-M3. Override with MINIMAX_BASE_URL for the mainland-China endpoint (https://api.minimaxi.com/v1).

Notes

• Temperature defaults to 1.0 because MiniMax requires temperature in (0.0, 1.0] and rejects 0.
• The module imports OpenAILLMService directly (rather than going through the generic autodiscovery helper) so it works on the project's declared minimum Python 3.10 — Pipecat has no native MiniMax LLM service, only a TTS one, so this is the idiomatic wiring.

Testing

• Verified the full path create_llm_service("minimax", ...) resolves to an OpenAILLMService with the correct base URL, default model MiniMax-M3, and temperature 1.0; model/base URL/temperature overrides are honored and missing-key validation fails early.
• Confirmed against the live MiniMax OpenAI-compatible endpoint that the MiniMax-M3 model id and auth are accepted.
• ruff check/ruff format clean on the new module.

[vercel]

Someone is attempting to deploy a commit to the akdeb's projects Team on Vercel.

A member of the Team first needs to authorize it.