feat: add MiniMax as LLM provider for agent evaluation

[octo-patch]

Summary

• Add MiniMax as a first-class LLM provider for agent evaluation alongside OpenAI, Anthropic, and Google
• Implement MiniMaxLMAgent and MiniMaxFeedbackAgent using MiniMax's OpenAI-compatible API (https://api.minimax.io/v1)
• Support MiniMax-M2.7 and MiniMax-M2.7-highspeed models with automatic temperature clamping

Changes

File	Description
mint/agents/minimax_lm_agent.py	Primary LM agent extending OpenAILMAgent with MiniMax API endpoint, MINIMAX_API_KEY env var, and temperature clamping
mint/agents/minimax_feedback_agent.py	Feedback agent for expert evaluation using MiniMax models
mint/agents/__init__.py	Export new MiniMax agent classes
mint/configs/config_variables.py	Add MiniMax-M2.7 and MiniMax-M2.7-highspeed to EVALUATED_MODEL_LIST
README.md	Add supported LLM providers table and MiniMax usage documentation
tests/	45 tests: 27 unit + 6 integration + 3 live API + 9 framework tests

Usage

export MINIMAX_API_KEY="your-api-key"
# MiniMax models are already configured in EVALUATED_MODEL_LIST

Test plan

• All 45 tests pass (including live API tests against MiniMax endpoint)
• Dynamic agent instantiation works via getattr(agents, "MiniMaxLMAgent") pattern
• Temperature clamping correctly maps 0.0 to 0.01 for MiniMax API compatibility
• Token usage tracking works with OpenAIObject-to-int conversion
• Feedback agent correctly handles textual and binary feedback forms