Add support for GPT-5 family and recent OpenAI o-series models

[cterdam]

Summary

• Add GPT-5 family models (gpt-5, gpt-5.1, gpt-5.2, gpt-5.4 and their mini/nano/pro/chat variants) to AVAILABLE_LLMS
• Add missing o-series models (o3, o3-pro, o4-mini) and other GPT-4 models (gpt-4.5-preview, gpt-4-turbo, chatgpt-4o-latest, gpt-3.5-turbo) to AVAILABLE_LLMS
• Fix get_response_from_llm and get_batch_responses_from_llm to use max_completion_tokens instead of max_tokens for GPT-5 models, since the latter raises Unsupported parameter: 'max_tokens' is not supported with this model
• Add o4 to the o-series routing in get_response_from_llm and create_client so o4-mini is handled correctly

Test plan

• Verify GPT-5 models use max_completion_tokens and complete successfully
• Verify older GPT-4 models still use max_tokens and are unaffected
• Verify o4-mini routes through the o-series code path (max_completion_tokens, no system message, temperature=1)
• Verify ensemble reviews (num_reviews_ensemble > 1) work with GPT-5 models via get_batch_responses_from_llm

🤖 Generated with Claude Code