feat: add MiniMax as a first-class provider

[octo-patch]

Reason: Add MiniMax as a first-class onboarding provider with MiniMax-M3 and MiniMax-M2.7 choices, global and CN endpoints, and both OpenAI-compatible and Anthropic-compatible routing.

Changes

• config.py: resolve MiniMax models through LiteLLM with a region-selected base URL. MINIMAX_REGION (global default, or cn) selects the endpoint set. The OpenAI-compatible route (e.g. minimax/MiniMax-M3) uses the /v1 base URL; the Anthropic-compatible route (e.g. anthropic/MiniMax-M3) uses the /anthropic base URL. MINIMAX_API_BASE / MINIMAX_ANTHROPIC_BASE allow explicit overrides, and MINIMAX_API_KEY is forwarded when set. Non-MiniMax resolution is unchanged.
• .env.example: document MINIMAX_API_KEY, MINIMAX_REGION, the optional base-URL overrides, and DEFAULT_MODEL examples for MiniMax-M3 and MiniMax-M2.7 on both routing styles.
• README.md: list MINIMAX_API_KEY in the required-provider setup section.
• tests/test_minimax_provider.py: unit tests covering model detection, region selection, endpoint resolution for both regions and both routing styles, override precedence, kwargs building, and .env.example documentation.

Checks

• python -m pytest tests/test_minimax_provider.py — 24 passed
• python -m py_compile config.py