Fix #180: Handle unsupported response_format parameter for non-OpenAI providers

[hkc5]

Problem

Ontology generation fails with 500 error when using LLM providers like Groq that don't support the OpenAI-specific response_format={"type": "json_object"} parameter.

Root Cause

The code uses response_format={"type": "json_object"} which is an OpenAI-specific feature. When using Groq or other providers that don't support this parameter, the API call fails with a 500 error.

Solution

Added graceful fallback logic in three files:

1. llm_client.py - Core LLM client with automatic retry without response_format
2. simulation_config_generator.py - Direct API calls with fallback
3. oasis_profile_generator.py - Direct API calls with fallback

When a 400/500 error related to response_format is detected, the code automatically retries without the parameter and relies on the system prompt to enforce JSON output format.

Testing

• Tested with Groq API (llama-3.1-8b-instant, llama-3.2-1b-preview, llama-3.1-70b-versatile)
• Backwards compatible with OpenAI API
• No breaking changes to existing functionality

Fixes #180

[lslsl3q]

很关键，希望赶紧修复

[lslsl3q]

感谢，已经验证确实有用，可以用，终于可以使用我自己的API了。

[hkc5]

Hi @666ghj @hzr1937,

This PR fixes the 500 error when using non-OpenAI LLM providers (like Groq). The issue was the hardcoded response_format={"type": "json_object"} which isn't supported by all providers.

The fix adds graceful fallback - when a provider doesn't support response_format, it retries without it while relying on the system prompt for JSON enforcement.

A user has already verified this works (see comments above). Would appreciate a review when you have time!

Thanks