#12477: fix(agents): prevent TimeoutOverflowWarning when timeout is disabled

## Summary `resolveAgentTimeoutMs` uses a 30-day constant (`2,592,000,000 ms`) to represent "no timeout" when the caller sets `timeout: 0`. This exceeds Node's 32-bit signed integer limit for `setTimeout` (`2,147,483,647 ms`), causing Node to emit `TimeoutOverflowWarning` and clamp the value to `1 ms` — which prematurely kills sub-agents almost immediately instead of letting them run indefinitely. The fix eliminates the overflowing constant entirely. Instead of scheduling a timer with a huge value, callers skip timer creation when no timeout is requested. ## Changes - `src/agents/timeout.ts` — return `0` for "no timeout" instead of the overflowing 30-day constant; export a documented `NO_TIMEOUT` sentinel callers can reference - `src/agents/pi-embedded-runner/run/attempt.ts` — skip the abort timer when `timeoutMs` is `0` instead of passing it to `setTimeout` - `src/gateway/server-methods/agent-job.ts` — skip the completion timer when `timeoutMs` is `0` `block-reply-pipeline.ts` already handles `timeoutMs <= 0` correctly (returns the promise directly without a timer). ## Test plan - `sessions_spawn` with `timeout: 0` no longer emits `TimeoutOverflowWarning` in Node logs - Sub-agents with no explicit timeout run until completion instead of being killed at 1 ms - Explicit positive timeouts (e.g. `timeout: 600`) behave exactly as before - `clearTimeout(undefined)` is a safe no-op in Node, so cleanup paths remain correct Fixes #12407 <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR changes the agent timeout semantics so that an explicit `timeout: 0` no longer maps to a huge “effectively infinite” millisecond value (which overflows Node’s `setTimeout` limit and gets clamped to `1ms`). Instead, `resolveAgentTimeoutMs` returns a `NO_TIMEOUT = 0` sentinel and call sites in the embedded runner and gateway job-waiter skip creating timers when `timeoutMs === 0`. The overall direction matches the described bug (prevent `TimeoutOverflowWarning` and premature aborts), but there’s one behavior mismatch in the gateway’s `waitForAgentJob` path that currently treats `timeoutMs <= 0` as an immediate timeout response rather than an indefinite wait. <h3>Confidence Score: 3/5</h3> - This PR addresses the Node timeout overflow issue but has a semantic regression in the gateway wait path for `timeoutMs=0`. - Main fix (stop scheduling overflowing timers) is correct in the embedded runner, but `src/gateway/server-methods/agent-job.ts` still returns immediately for `timeoutMs <= 0`, which conflicts with the new `NO_TIMEOUT = 0` sentinel meaning and will cause `agent.wait` to report an immediate timeout when clients pass 0 to mean “no timeout”. - src/gateway/server-methods/agent-job.ts <!-- greptile_other_comments_section --> _{(5/5) You can turn off certain types of comments like style [here](https://app.greptile.com/review/github)!} <!-- /greptile_comment -->