Allow longer OpenAI Responses requests before timing out

[susano]

In 0.32a3, OpenAI Responses SDK times out by default after 10 minutes and retries twice, leading to 30 minutes wait before time out error for long API calls (e.g. gpt-5.5-pro xhigh with max_tokens > 25,000).

Suggested fix is introducing --timeout and --max-retries model options for Response models. Example agent implementation that works for me is at 6519b1a

Details

I'm using llm 0.32a3 with an extra OpenAI model configured for gpt-5.5-pro:

- model_id: gpt-5.5-pro
  model_name: gpt-5.5-pro
  responses: true
  reasoning: true
  vision: true
  supports_tools: true
  supports_schema: true
  can_stream: false

I'm trying to run it with a template that gives it a large output budget:

model: gpt-5.5-pro
options:
  reasoning_effort: xhigh
  max_tokens: 25000

This currently works for me with max_tokens: 12500, but 25000 and above consistently time out after what seems to be exactly 30 minutes:

cat prompt | llm -t gpt-pro-max > out.pro
Error: Request timed out.

After adding model --timeout and --max-retries options (6519b1a) and changing the template to the following, I have had a successful 17 minutes command run.

model: gpt-5.5-pro
options:
  reasoning_effort: xhigh
  max_tokens: 50000
  max_retries: 0
  timeout: 7200 # two hours