Add MiniMax Cloud API as LLM provider

[octo-patch]

Summary

Add MiniMax as a first-class LLM provider via a new minimax model format.

• MiniMaxAPI(GPTAPI) subclass that overrides generate_request_data() and _stream_chat() to handle MiniMax model types not recognised by the base class's routing logic, and to fix streaming response parsing (the base class assumes finish_reason is always present in every SSE chunk, but MiniMax—following standard OpenAI behaviour—omits it in non-final chunks)
• minimax preset dict with sensible defaults (MiniMax-M1, configurable via MINIMAX_MODEL / MINIMAX_API_KEY env vars)
• Documentation updates in README.md and README_zh-CN.md listing minimax as a --model_format option
• .env.example updated with MINIMAX_API_KEY and MINIMAX_MODEL placeholders

Usage

export MINIMAX_API_KEY="your-key"
python -m mindsearch.app --lang en --model_format minimax --search_engine DuckDuckGoSearch

Optionally override the model:

export MINIMAX_MODEL="MiniMax-M1-highspeed"

Test plan

• 16 unit tests covering preset config, MiniMaxAPI request generation, parameter stripping, env var overrides, and .env.example validation — all pass
• 3 integration tests (basic chat, streaming, init_agent resolution) — all pass with a valid MINIMAX_API_KEY
• Run with python -m unittest tests.test_minimax_models tests.test_minimax_integration