feat: add MiniMax model compatibility support

[octo-patch]

Summary

• Add automatic MiniMax model detection via model name and base URL
• Handle MiniMax API constraints: clamp temperature to 0.01 when set to 0 (MiniMax requires temperature ∈ (0.0, 1.0]), and replace unsupported response_format with prompt-based JSON instruction injection
• Strip <think> tags from MiniMax M2.5 model responses in simulation_config_generator.py and oasis_profile_generator.py
• Add MiniMax configuration examples to .env.example and documentation to README.md / README-EN.md
• Add 23 unit tests for all MiniMax compatibility functions

Changes

File	Description
backend/app/utils/llm_client.py	Add _is_minimax(), _clamp_temperature(), _inject_json_instruction() helpers; update LLMClient.chat() to conditionally skip response_format for MiniMax
backend/app/services/simulation_config_generator.py	Use MiniMax-aware JSON parsing and <think> tag stripping in _call_llm_with_retry()
backend/app/services/oasis_profile_generator.py	Use MiniMax-aware JSON parsing and <think> tag stripping in profile generation
.env.example	Add commented MiniMax configuration section
README.md	Add collapsible MiniMax setup guide (Chinese)
README-EN.md	Add collapsible MiniMax setup guide (English)
backend/tests/test_minimax_compat.py	23 unit tests covering detection, temperature clamping, JSON injection, and response parsing

Test plan

• All 23 unit tests pass (pytest backend/tests/test_minimax_compat.py)
• Verify MiniMax M2.5 model works end-to-end with simulation (requires MiniMax API key)
• Verify existing non-MiniMax models (e.g., qwen-plus) are unaffected

MiniMax Configuration

LLM_API_KEY=your_minimax_api_key
LLM_BASE_URL=https://api.minimax.io/v1
LLM_MODEL_NAME=MiniMax-M2.5

Supported models: MiniMax-M2.5 (flagship, 204K context) and MiniMax-M2.5-highspeed (faster variant).

🤖 Generated with Claude Code