Add MiniMax as a built-in LLM provider

[octo-patch]

Summary

• Add MiniMax M2.7 / M2.7-highspeed as a third language model backend alongside ChatGPT and LLaMA-2
• New MiniMax class extending AbstractLanguageModel, using the OpenAI-compatible API at https://api.minimax.io/v1
• Temperature clamping to [0, 1], per-call response handling (MiniMax does not support the n parameter), and cumulative token/cost tracking

Changes

File	Description
graph_of_thoughts/language_models/minimax.py	New MiniMax class with query(), chat(), get_response_texts()
graph_of_thoughts/language_models/init.py	Export MiniMax
graph_of_thoughts/language_models/config_template.json	Add minimax and minimax-highspeed config entries
graph_of_thoughts/language_models/README.md	MiniMax setup documentation with config table
README.md	Quick Start code comments showing MiniMax usage
tests/test_minimax.py	21 unit tests (init, query, caching, token tracking, config template)
tests/test_minimax_integration.py	3 integration tests (simple query, multiple responses, highspeed model)

Usage

from graph_of_thoughts import language_models

# Set MINIMAX_API_KEY env var or fill api_key in config.json
lm = language_models.MiniMax("config.json", model_name="minimax")

Test plan

• 21 unit tests pass (mocked API calls)
• 3 integration tests pass (real MiniMax API calls with MINIMAX_API_KEY)
• Verify existing ChatGPT and LLaMA tests are unaffected