Add exp_auto_restart plugin

[bbassie]

Adds exp_auto_restart (@expcluster/auto-restart), a controller-only plugin that restarts hosts, and optionally the controller, which report "Restart Required" once no players have been online for a configurable time.

How it works. Every 10 s it walks the controller's system info. A connected host with restartRequired and canRestart whose running instances have had no players for exp_auto_restart.idle_seconds gets HostRestartRequest, the same request the web UI button sends. Players are counted from the users online on instances with status running, so stale online sets after a host crash do not hold anything back. When the host reconnects and reports its instances, the ones that were running are started again; instances with instance.auto_start are left to the host. Each host process is asked at most once, keyed on its process start time, so a stale system info poll cannot trigger a second restart. A refusal from the host (no --can-restart, downgrade) is logged once and not retried, transport errors retry on the next check.

Config, all on the controller: idle_seconds (300), scope (host or cluster, whether players on other hosts also hold back a restart), start_instances (true), restart_controller (false, restarts the controller when the whole cluster is idle, after the same downgrade check the web UI does).

It uses the function-style default export that clusterio master now documents, so it loads without the "deprecated class export" warning the other plugins get.

Testing. 53 tap assertions around a real Controller with sendTo stubbed: idle tracking, the idle threshold, once per process, canRestart false, cluster scope, refusal versus transport errors, instance restart after reconnect, pending expiry, and the controller path. Smoke tested in a standalone dev cluster with a host in a restart loop and a Factorio 2.1.19 instance: after a restart-required config change the host was restarted once the idle time passed, came back in 2 s and the instance was started again, with no second restart.

Notes.

• Clusterio does not yet flag a restart after a plugin install, only after updates and restart-flagged config changes. Mark restart required after installing a new plugin clusterio/clusterio#1042 fixes that. Until it lands this plugin will not react to plugin installs, the README says so.
• Unrelated to this PR: main does not build against clusterio master since clusterio/clusterio@7122cbcc (typed hook system). exp_roles web/index.tsx fails on PluginExtensionProps and exp_scenario on controller.plugins, so Node CI will be red until that is caught up.

🤖 Generated with Claude Code