fix(config): honor configured timeout for auxiliary title generation (#32729)

[rodboev]

Summary

Users running slow local LLMs see "Auxiliary title generation failed: Request timed out." after every session, even after setting auxiliary.title_generation.timeout to a large value in config.yaml. The configured timeout is silently ignored.

The root cause is in agent/title_generator.py. generate_title() declares timeout: float = 30.0 as a default parameter (line 32). auto_title_session() calls it without passing timeout (line 99), so the parameter takes its hardcoded default of 30.0. This value is forwarded to call_llm(task="title_generation", ..., timeout=30.0). Inside call_llm() at agent/auxiliary_client.py:5027, the resolution logic is effective_timeout = timeout if timeout is not None else _get_task_timeout(task). Because timeout is 30.0 (not None), the _get_task_timeout("title_generation") branch, which reads auxiliary.title_generation.timeout from config.yaml, is never reached. The config value is dead code for title generation.

The fix changes the timeout default from 30.0 to None. With timeout=None, call_llm falls through to _get_task_timeout("title_generation"), which reads the config. When the config key is absent, _get_task_timeout falls back to _DEFAULT_AUX_TIMEOUT (30 seconds), preserving the existing behavior for users who have not set the config. Users who have set it now get the value they configured.

Changes

• agent/title_generator.py: change generate_title() signature from timeout: float = 30.0 to timeout: Optional[float] = None (+1 line changed)
• tests/agent/test_title_generator.py: add test_timeout_defaults_to_none and test_explicit_timeout_passed_through to TestGenerateTitle (+~20 lines)

Validation

Scenario	Before	After
auxiliary.title_generation.timeout: 1800 in config.yaml	ignored, hardcoded 30s timeout	honored, 1800s timeout
No auxiliary.title_generation.timeout in config.yaml	30s timeout (from hardcoded default)	30s timeout (from _DEFAULT_AUX_TIMEOUT fallback, unchanged)
Caller passes timeout=120.0 explicitly	120s timeout	120s timeout (unchanged)
Fast cloud model, no config override	title generates within 30s	title generates within 30s (unchanged)
Slow local model (e.g. qwen3.6:27b on limited VRAM)	"Auxiliary title generation failed: Request timed out."	succeeds if within configured timeout

Test plan

• pytest tests/agent/test_title_generator.py -v --timeout=0 — 22 passed
• New: test_timeout_defaults_to_none asserts call_llm receives timeout=None when no timeout is passed to generate_title()
• New: test_explicit_timeout_passed_through asserts call_llm receives timeout=120.0 when generate_title(..., timeout=120.0) is called

Not in scope

Other auxiliary tasks may have similar hardcoded-default patterns (compression, vision, web_extract, session_search). This PR fixes only title generation, which is the one reported upstream. A broader audit of all auxiliary task timeout defaults would be a follow-up.

Upstream

Closes #32729.
Reported by @veryheavypickle.

[kshitijk4poor]

Closing as superseded by #56322 (salvage of #41844) for #32729. Your title-generator one-liner + tests are correct, but this PR also bundles an unrelated arm64 Docker CI rewrite (.github/workflows/docker.yml) that isn't part of the timeout bug — so it can't land as-is for this issue. #41844 carries just the focused fix. If the Docker CI change has independent merit, please resubmit it as its own PR — happy to look at it separately. Credit for the fix.