Hot reload — reload application code without dropping connections

[EdmondDantes]

Motivation

The server is long-lived: handler/bootloader code stays resident in each worker, so today any change to the PHP app requires a full process restart — which drops every in-flight and idle connection. We want to reload the application code in place, keeping listeners and live connections alive.

Scope

• Trigger: file-watcher (inotify) for dev; a signal (SIGHUP) and/or an admin API for prod.
• Reload: per-worker rolling reload — a worker stops accepting, drains in-flight requests via the existing shutdown-drain machinery (Server shutdown disposes server_scope while per-request coroutines are still in-flight #74), re-runs the bootloader (HttpServerConfig::setBootloader) to re-register handlers, resets opcache, then resumes. Other workers keep serving throughout → zero downtime.
• Preserve across reload: the listener sockets (don't rebind), and long-lived streams (WebSocket / SSE) — either drained with a 1001 Going Away / stream close, or handed to the new code where safe.

Open questions

• opcache invalidation strategy (full reset vs per-file).
• What worker state is allowed to survive a reload (bootloader-owned singletons, pools).
• Interaction with setWorkers() / the reactor pool.

Brief for now — filing so it isn't lost.