Make /api/shutdown actually exit the server process

[MasonXon]

Problem

restart (and stop-service) left a leftover csghub-lite serve process behind each time, so processes accumulated.

Root cause: handleShutdown only called s.http.Shutdown(), which closes the HTTP listeners. But the Run loop is blocked on <-ctx.Done() waiting for a signal that never comes via the HTTP path, so the process never exits. stopBackgroundService polls the health endpoint, sees it go unresponsive, assumes the stop succeeded, and starts a fresh server — while the old process keeps running.

Fix

Store the signal.NotifyContext cancel function on the Server and invoke it from handleShutdown after closing the listeners. This drives Run through the same exit path as SIGINT/SIGTERM (graceful engine + listener shutdown), so the process actually terminates.

• http.Server.Shutdown is idempotent (called once here, once in the Run exit path) — safe.
• shutdownRuntime (closeAllEngines + CloseAll) is idempotent — safe to run twice.
• shutdownCancel is nil-guarded; only set in Run, which always runs before the route can be hit.

Note

This prevents new orphans from accumulating. Existing leftover processes from before this fix are not auto-cleaned; clear them once with pkill -f "csghub-lite serve".

Before:

After: