feat: add MiniMax as first-class LLM provider

[octo-patch]

Summary

Add MiniMax as a dedicated LLM backend adapter alongside OpenAI, Anthropic, and Gemini. MiniMax provides an OpenAI-compatible API with powerful models:

• MiniMax-M2.7 — latest flagship model with 1M context window
• MiniMax-M2.7-highspeed — speed-optimized variant
• MiniMax-M2.5 — previous generation, 204K context
• MiniMax-M2.5-highspeed — speed-optimized, 204K context

Changes

• New adapter: src/core/component/llm/llm_adapter/minimax_adapter.py — MiniMaxAdapter extending LLMBackendAdapter via OpenAI SDK, with:
  • Temperature clamping to [0.01, 1.0] range for MiniMax API compatibility
  • <think>...</think> tag stripping for reasoning model responses
  • MINIMAX_API_KEY environment variable auto-detection
• Factory routing: Updated OpenAICompatibleClient to route provider: minimax to MiniMaxAdapter
• Backend config: Added minimax entry in llm_backends.yaml with 4 models
• Env template: Added MINIMAX_API_KEY section in env.template
• Documentation: Updated README.md and README.zh.md to list MiniMax as a supported backend
• Tests: 35 tests (32 unit + 3 integration) covering initialization, temperature clamping, think-tag stripping, streaming, error handling, and live API calls

Test plan

• 32 unit tests pass with mocked API calls
• 3 integration tests pass against live MiniMax API
• Verify minimax backend is properly configured in llm_backends.yaml
• Verify OpenAICompatibleClient routes to MiniMaxAdapter for provider: minimax
• Set MINIMAX_API_KEY and select minimax as default_backend in llm_backends.yaml, run the application

7 files changed, 665 additions