feat: add MiniMax as first-class LLM provider

[octo-patch]

Summary

Add MiniMax M2.7 and M2.7-highspeed as first-class LLM providers across both the main LLM module and the treesearch backend.

Changes

• ai_scientist/llm.py: Add MiniMax models to AVAILABLE_LLMS, create_client(), get_response_from_llm(), get_batch_responses_from_llm(), and make_llm_call() with temperature clamping (0.0, 1.0] and think-tag stripping
• ai_scientist/treesearch/backend/init.py: Add MiniMax temperature clamping and think-tag stripping in the backend query router
• ai_scientist/treesearch/backend/backend_openai.py: Add MiniMax client creation via MINIMAX_API_KEY env var and api.minimax.io base URL
• README.md: Add MiniMax model documentation and API key setup instructions
• tests/: Add 31 unit tests and 5 integration tests covering model detection, temperature clamping, think-tag stripping, client creation, and live API calls

MiniMax Models

Model	Context Window	Description
MiniMax-M2.7	204K tokens	Latest MiniMax model
MiniMax-M2.7-highspeed	204K tokens	Optimized for speed

Key Implementation Details

• Uses OpenAI-compatible API via openai.OpenAI(base_url="https://api.minimax.io/v1")
• Temperature automatically clamped to (0.0, 1.0] range required by MiniMax API
• think tags stripped from M2.7 reasoning responses
• No new dependencies required - reuses existing openai SDK

Test Plan

• 31 unit tests pass (model detection, temperature clamping, think-tag stripping, client creation, mocked API calls)
• 4 integration tests pass with live MiniMax API (client creation, M2.7 response, M2.7-highspeed response)
• Existing code paths unaffected - MiniMax branch is additive