Add MiniMax as LLM provider with M2.7 and M2.5 models

[octo-patch]

Summary

• Add MiniMax as an LLM provider in WikiChat via the OpenAI-compatible API
• Add MiniMax-M2.7, M2.7-highspeed, M2.5, and M2.5-highspeed models to llm_config.yaml using LiteLLM's openai/ prefix for routing
• Add MiniMax models to the Chainlit web UI model selector dropdown
• Update README.md to mention MiniMax in supported providers and API key setup

Details

MiniMax provides an OpenAI-compatible API at https://api.minimax.io/v1, making integration seamless through WikiChat's existing ChainLite/LiteLLM stack. No code changes are needed beyond configuration — once the MINIMAX_API_KEY environment variable is set, all pipeline stages (query, generation, filtering, drafting, refining, reranking) automatically work with MiniMax models.

Models added:

Engine name	Model	Context window
minimax-m27	MiniMax-M2.7	1M tokens
minimax-m27-highspeed	MiniMax-M2.7-highspeed	1M tokens
minimax-m25	MiniMax-M2.5	204K tokens
minimax-m25-highspeed	MiniMax-M2.5-highspeed	204K tokens

Usage:

# Set API key
echo "MINIMAX_API_KEY=your-key" >> API_KEYS

# Run WikiChat with MiniMax
inv demo --engine minimax-m27

Test plan

• 18 unit tests validating config structure, engine name uniqueness, LiteLLM prefix, UI dropdown, README mentions, engine resolution, and temperature handling
• 3 integration tests (MiniMax-M2.7 completion, M2.5-highspeed completion, streaming) — all passing
• All 21 tests pass: pytest tests/test_minimax.py -v