feat: add MiniMax as a first-class LLM provider

[octo-patch]

Summary

Add MiniMax as a first-class LLM provider in MemOS, following the same clean pattern as the existing Qwen and DeepSeek integrations.

What is MiniMax?

MiniMax is an AI company that provides OpenAI-compatible LLM APIs. Their latest models include:

• MiniMax-M2.5 — general-purpose model with 204K context window
• MiniMax-M2.5-highspeed — faster variant for latency-sensitive workloads

API Base URL: https://api.minimax.io/v1 (fully OpenAI-compatible)

Changes

• New file: src/memos/llms/minimax.py — MinimaxLLM class inheriting from OpenAILLM
• Config: MinimaxLLMConfig in src/memos/configs/llm.py with api_key, api_base, extra_body fields
• Factory: Registered minimax backend in both LLMFactory and LLMConfigFactory
• API Config: Added minimax_config() to APIConfig with environment variable support (MINIMAX_API_KEY, MINIMAX_API_BASE)
• Backend mapping: Added minimax to backend_model dicts in get_product_default_config() and create_user_config()
• Examples: Added MiniMax usage scenario (Scenario 7) in examples/basic_modules/llm.py
• Tests: Unit tests for config validation, generate, and streaming in tests/llms/test_minimax.py and tests/configs/test_llm.py
• Docs: Updated .env.example and README.md to list MiniMax as a supported provider

Usage

from memos.configs.llm import LLMConfigFactory
from memos.llms.factory import LLMFactory

config = LLMConfigFactory.model_validate({
    "backend": "minimax",
    "config": {
        "model_name_or_path": "MiniMax-M2.5",
        "api_key": "your-api-key",
        "api_base": "https://api.minimax.io/v1",
        "temperature": 0.7,
        "max_tokens": 1024,
    },
})
llm = LLMFactory.from_config(config)
response = llm.generate([{"role": "user", "content": "Hello!"}])

Or via environment variables:

MOS_CHAT_MODEL_PROVIDER=minimax
MOS_CHAT_MODEL=MiniMax-M2.5
MINIMAX_API_KEY=your-api-key

Test Plan

• Unit tests pass: pytest tests/llms/test_minimax.py (4 tests)
• Config tests pass: pytest tests/configs/test_llm.py (6 tests)
• Existing tests unaffected: pytest tests/llms/test_deepseek.py tests/llms/test_qwen.py (4 tests)
• Integration tested with real MiniMax API (both streaming and non-streaming)