[Fix #3228] idle-timeout not working in cluster mode (#3235)

[MitchLewis930]

PR_057

---

Note

Medium Risk
Touches server accept-loop shutdown behavior and introduces a new worker-to-master SIGTERM path, which could affect shutdown/restart semantics if idle timeouts are misdetected.

Overview
Fixes idle_timeout behavior in cluster mode by having Puma::Server record when an idle timeout stops the accept loop (idle_timeout_reached), and having cluster workers propagate that event by sending SIGTERM to the master so the whole cluster shuts down.

Adds integration coverage for idle_timeout in both single and cluster modes, and includes a small typo fix in a comment in runner.rb.

^{Written by Cursor Bugbot for commit 8d1f77b. This will update automatically on new commits. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{Bugbot Autofix is OFF. To automatically fix reported issues with Cloud Agents, enable Autofix in the Cursor dashboard.}

[cursor]

Missing exception handling for dead master process signal

Low Severity

The Process.kill "SIGTERM", master call lacks exception handling for Errno::ESRCH, which is raised when the target process no longer exists. If the master process dies between the check_pipe detection and this line (a narrow race window), an unhandled exception occurs. The rest of the codebase consistently handles this exception for Process.kill calls (e.g., in worker_handle.rb). Adding rescue Errno::ESRCH would align with existing patterns.