Add MiniMax extraction backend

[octo-patch]

Reason: Add MiniMax-M3 and MiniMax-M2.7 support to the OpenAI-compatible extraction backend with regional endpoint selection.

Changes

• Add the MiniMax backend with global and configurable regional OpenAI-compatible endpoints.
• Support MiniMax-M3 by default and MiniMax-M2.7 through the model override.
• Enable automatic API-key detection and image-capable extraction for the backend.
• Document the backend environment variables and CLI usage.

Checks

• pytest -q tests/test_llm_backends.py tests/test_provider_registry.py tests/test_image_vision.py (127 passed)
• ruff check graphify/llm.py tests/test_llm_backends.py