feat: route MiniMax through the Anthropic-compatible endpoint and add test coverage

[octo-patch]

Reason: Refresh the MiniMax integration to cover the Anthropic-compatible endpoints and add MiniMax test coverage.

What changed

• server.py: MiniMax models (minimax/ prefix) are now routed through LiteLLM's anthropic/ provider to the MiniMax Anthropic-compatible endpoint (https://api.minimax.io/anthropic, China region https://api.minimaxi.com/anthropic), matching the target endpoint configuration. The OpenAI-compatible endpoint remains available via MINIMAX_API_STYLE=openai with MINIMAX_BASE_URL (https://api.minimax.io/v1 / https://api.minimaxi.com/v1).
• Added MINIMAX_API_KEY, MINIMAX_BASE_URL, MINIMAX_ANTHROPIC_BASE_URL, and MINIMAX_API_STYLE configuration.
• Added MINIMAX_MODELS = ["MiniMax-M3", "MiniMax-M2.7"] with model prefixing for bare MiniMax model names, and PREFERRED_PROVIDER=minimax support for haiku/sonnet mapping.
• Thinking blocks are forwarded for MiniMax models routed through the Anthropic-compatible endpoint.
• tests.py: Added a proxy-only MiniMax scenario (minimax/MiniMax-M3) with a test_proxy_only harness function that validates the Anthropic-format response structure directly, since there is no native Anthropic equivalent to compare against.
• .env.example / README.md: Documented the new MiniMax configuration.

Checks run

• python3 -m py_compile server.py tests.py
• python3 tests.py --help
• Exercised the model-mapping and conversion logic against a minimal LiteLLM stub (prefixing, minimax/ -> anthropic/ routing, thinking forwarding, base-URL selection).