fix(title_gen): honor auxiliary.title_generation.timeout config setting (#41812)

[Elshayib]

Summary

Title generation ignored the auxiliary.title_generation.timeout config value, always using a hardcoded 30s default. Users with slow local models (e.g. Ollama on CPU) who set auxiliary.title_generation.timeout: 300 still saw timeouts at 30s.

Root Cause

The call chain maybe_auto_title() → auto_title_session() → generate_title(timeout=30.0) → call_llm(timeout=30.0) never read the config value. call_llm() already supports timeout=None → _get_task_timeout(task) which reads auxiliary.<task>.timeout, but the hardcoded 30.0 default overrode it.

Changes

• agent/title_generator.py: Changed generate_title() default from timeout: float = 30.0 to timeout: Optional[float] = None. Thread timeout through maybe_auto_title() and auto_title_session().
• gateway/run.py: Read auxiliary.title_generation.timeout via _get_task_timeout("title_generation") and pass it as timeout to maybe_auto_title().
• tests/agent/test_title_generator.py: 5 new tests covering default-None behavior, explicit timeout forwarding through all three layers.

Validation

• 25/25 title generator tests pass
• 209/209 auxiliary client tests pass

Fixes #41812

[Elshayib]

CI failure is a pre-existing flaky test (test_concurrent_handles_tool_error — race condition in concurrent tool execution under CI load). Passes locally. Not related to this change. Requesting re-run or review override.

[kshitijk4poor]

Closing as superseded by #56322 (salvage of #41844) for #32729. Both fix the hardcoded title-generation timeout, but #41841 threads a timeout kwarg through auto_title_session/maybe_auto_title and re-computes _get_task_timeout('title_generation') in gateway/run.py — which duplicates the resolution call_llm already does internally when timeout=None. #41844 achieves the same result with the one-line default change and no gateway plumbing. Credit for the correct fix. Feel free to push back.