feat: add MiniMax as alternative LLM provider for prompt optimization

[octo-patch]

Summary

Add MiniMax as an alternative LLM provider for prompt optimization alongside the existing Zhipu AI backend.

• Provider preset system in prompt_optimize.py: --provider minimax|zhipu|openai for easy switching, with MINIMAX_API_KEY env var auto-detection
• Gradio demo update: LLM Provider dropdown in the web UI, configurable via LLM_PROVIDER env var
• MiniMax M2.7 model with temperature clamping for API compatibility
• README/README_zh updated with MiniMax usage examples
• 18 unit tests + 3 integration tests covering provider presets, temperature handling, message construction, and end-to-end API calls

Changed files (7 files, 444 additions)

File	Change
inference/prompt_optimize.py	Added PROVIDER_PRESETS dict, --provider arg, MINIMAX_API_KEY env var support
inference/gradio_web_demo.py	Added provider dropdown, LLM_PROVIDER env var, imported PROVIDER_PRESETS
README.md	Added MiniMax usage examples for prompt optimization
README_zh.md	Added MiniMax usage examples (Chinese)
tests/__init__.py	Test package init
tests/test_prompt_optimize.py	18 unit tests
tests/test_integration_minimax.py	3 integration tests

Test plan

• All 18 unit tests pass (python -m unittest tests.test_prompt_optimize -v)
• All 3 integration tests pass with real MiniMax API (MINIMAX_API_KEY=... python -m unittest tests.test_integration_minimax -v)
• Existing prompt_optimize.py CLI behavior unchanged (defaults to zhipu provider)
• Verify Gradio demo with LLM_PROVIDER=minimax (requires GPU for diffusion model)

---

This PR adds MiniMax as an alternative — the default behavior (Zhipu AI) is fully preserved and no existing functionality is changed.