feat: add MiniMax as alternative LLM provider for evaluation

[octo-patch]

Summary

• Add MiniMax as an alternative LLM provider for GPT-score evaluation alongside OpenAI
• Make GPTEvaluation class configurable with --provider and --model CLI arguments
• Auto-detect provider from environment variables (MINIMAX_API_KEY / OPENAI_API_KEY)
• Handle MiniMax-specific constraints: temperature clamping (0.0, 1.0], <think> tag stripping for M2.7 models
• Replace hardcoded API key placeholder with proper env var support

Changes

File	Change
challenge/gpt_eval.py	Multi-provider support, PROVIDER_CONFIGS, auto-detection, temperature clamping, think-tag stripping
challenge/evaluation.py	--provider and --model CLI arguments, pass through to GPTEvaluation
challenge/tests/test_gpt_eval.py	31 unit tests covering provider resolution, temperature clamping, think-tag stripping, config validation
challenge/tests/test_integration.py	3 MiniMax + 1 OpenAI live API integration tests (skipped if key unavailable)
README.md	LLM provider configuration documentation with usage examples

MiniMax API Reference

• Chat (OpenAI Compatible): https://platform.minimax.io/docs/api-reference/text-openai-api
• Models: MiniMax-M2.7, MiniMax-M2.7-highspeed
• Base URL: https://api.minimax.io/v1

Usage

# Using MiniMax for evaluation
export MINIMAX_API_KEY="your-key"
python challenge/evaluation.py --root_path1 pred.json --root_path2 test.json --provider minimax

# Using OpenAI (default, backward compatible)
export OPENAI_API_KEY="your-key"
python challenge/evaluation.py --root_path1 pred.json --root_path2 test.json

Test plan

• 31 unit tests pass (pytest challenge/tests/test_gpt_eval.py)
• 3 MiniMax integration tests pass with live API
• Backward compatible — existing OpenAI usage unchanged when no new args are provided