Add MiniMax as an LLM provider

[octo-patch]

Reason: ChatLLMFactory only accepted a fixed provider list and rejected every other identifier, so MiniMax could not be selected.

What changed

ChatLLMFactory previously hard-coded its provider list and raised ValueError for any identifier outside it. This PR adds MiniMax as a first-class provider following the existing per-provider pattern.

• New llm/minimax_chat.py exposing AssistantChatMiniMax (OpenAI-compatible route) and AssistantChatMiniMaxAnthropic (Anthropic-compatible route), plus create_minimax_chat and get_minimax_models, mirroring the existing openai_chat.py / anthropic_chat.py modules.
  • Both global (api.minimax.io) and China (api.minimaxi.com) routes are supported through a region setting (global_en / cn_zh), each providing an OpenAI-compatible (/v1) and an Anthropic-compatible (/anthropic) base URL.
  • An api setting (openai / anthropic) selects the compatibility route.
  • get_minimax_models discovers models over the OpenAI-compatible route and falls back to the current release list (MiniMax-M3, MiniMax-M2.7).
  • Credentials are read from the MINIMAX_API_KEY environment variable.
• llm/llm_factory.py registers minimax in get_valid_providers, get_valid_models, the get_chat_model dispatch, and the return type union.
• New configs/minimax.yaml provider config modeled on the existing provider configs, defaulting to MiniMax-M3 on the global_en OpenAI-compatible route.
• New tests/unittests/llm/test_minimax_provider.py verifying provider registration, the regional endpoint map, and the current model list. The module is guarded with pytest.importorskip for the optional runtime dependencies.

Checks

• python -m py_compile on the changed modules and tests — passed.
• python -m isort --check-only on the changed files — passed.
• python -m pytest tests/unittests/llm/test_minimax_provider.py — collected; skipped locally because the optional LLM runtime dependencies are not installed in this environment.

[kapelame]

Thanks for the review. I formatted the MiniMax chat provider to satisfy the lint configuration and pushed a new commit. I ran Black, Ruff, isort, and the MiniMax provider test (skipped because optional runtime dependencies are unavailable locally).

[kapelame]

Thanks for the review. I formatted the MiniMax chat provider to satisfy the lint configuration and pushed a new commit. I ran Black, Ruff, isort, and the MiniMax provider test (skipped because optional runtime dependencies are unavailable locally).