fix: add request timeout to prevent infinite hangs during scoring loop

[TristanSchneider-dev]

Add request timeout to prevent infinite hangs during scoring loop

Midway through the pipeline, I noticed that it got stuck indefinitely because the API stopped responding. :(

Problem

The pipeline can hang indefinitely when the OpenAI API server stops responding midrequest.

Error flow

During the detailed requirement scoring loop in minimal_agent.py, every requirement triggers a separate API call. Earlier logs show a 503 Service Unavailable that eventually succeeded after retry. At some point the API connection stalled silently ? the traceback shows httpcore blocked on self._protocol.read_event.wait() with no timeout configured on ChatOpenAI. The asyncio.exceptions.CancelledError only appeared after a manual KeyboardInterrupt; without that the process hangs forever.

Fix

Added timeout and max_retries to the ChatOpenAI constructor in treesearch/llm/query.py:

_cfg = get_config()
model = ChatOpenAI(
    model=self._model,
    temperature=self._temperature,
    use_responses_api=True,
    timeout=_cfg.agent.code.request_timeout,   # 120 seconds
    max_retries=_cfg.agent.code.max_retries,    # 3 retries
)

Parameters are reused from config.toml

Parameter	Value	Description
request_timeout	120s	Timeout per API call
max_retries	3	Automatic retry attempts

Why it works..

A x-second timeout ensures unresponsive calls raise a httpx.TimeoutException instead of hanging forever.
The exception is caught by the existing try/except blocks in minimal_agent.py, which mark the requirement as is_fulfilled=False and continue the pipeline gracefully.
3 automatic retries handle transient failures (like the observed 503) without manual intervention.