feat: add MiniMax Cloud API as engine backend

[octo-patch]

Summary

Add MiniMax as a third engine type alongside HuggingFace and vLLM, enabling users to serve MiniMax cloud models through the same OpenAI-compatible API without local GPU.

Changes

• New engine: api/engine/minimax.py — MiniMaxEngine class using OpenAI SDK with MiniMax's OpenAI-compatible endpoint
• New routes: api/minimax_routes/ — Chat completion and text completion endpoints for MiniMax
• Config: MINIMAX_API_KEY, MINIMAX_API_BASE environment variables via MiniMaxSettings
• Temperature clamping: Automatically clamps temperature to MiniMax's accepted [0, 1] range
• Streaming support: Both streaming and non-streaming responses
• README: Added MiniMax section with setup instructions and supported models table
• Tests: 29 unit tests + 3 integration tests

Supported Models

Model	Context Length
MiniMax-M2.7	1M
MiniMax-M2.7-highspeed	1M
MiniMax-M2.5	245K
MiniMax-M2.5-highspeed	204K

Usage

export ENGINE=minimax
export MODEL_NAME=MiniMax-M2.7
export MINIMAX_API_KEY=your-api-key
python -m api.server

The API is fully compatible — existing client code using OpenAI SDK works without modification.

Test Plan

• 29 unit tests covering engine initialization, temperature clamping, config, route kwargs building, module imports
• 3 integration tests with real MiniMax API (chat completion, streaming, temperature clamping)
• All tests passing

---

11 files changed, 769 additions(+), 5 deletions(-)