feat: add MiniMax provider configurations
Reason: add MiniMax provider configurations with MiniMax-M3.1-Flash as the default text model and MiniMax-M3 as the other predefined choice.
Adds global and China configurations for the OpenAI-compatible and Anthropic-compatible endpoints, shared API key handling, endpoint overrides, and Docker environment defaults. The current model list is used by both agent and planner pickers and by the provider factory.
Saved and custom model IDs remain usable, including legacy IDs no longer shown as ordinary choices. Loading settings restores each model together with its provider's current choices; only a user provider change resets the model selection. Existing message content continues through the SDK unchanged except for its normal protocol conversion.
Validation passed in an isolated Python 3.11 environment with the repository-pinned SDK adapters and Gradio version:
.venv/bin/python -m pytest -q tests/test_minimax_provider.py: 73 passed. Offline tests capture serialized synchronous and asynchronous HTTP requests for both regions/protocols, default and explicit models, agent/research initialization, saved agent/planner restoration (including the real planner settings consumer), endpoint/key overrides, and multimodal message serialization..venv/bin/python -m ruff check --select E9,F63,F7,F82 src/utils/config.py src/utils/llm_provider.py src/webui/components/agent_settings_tab.py src/webui/webui_manager.py tests/test_minimax_provider.py.venv/bin/python -m compileall -q src/utils/config.py src/utils/llm_provider.py src/webui/components/agent_settings_tab.py src/webui/webui_manager.py tests/test_minimax_provider.pydocker compose config -qgit diff --check
Validation used the actual application imports and a targeted SDK/Gradio environment, without the unused agent-memory backend or browser binaries. Pytest emitted Gradio dropdown-construction warnings and a websockets deprecation warning. No paid inference, live browser task, or full Docker image build has been performed.