fix(desktop): wait for backend exit before reloading on connection-config apply

[AhmetArif0]

Problem

hermes:connection-config:apply called resetHermesConnection() (which sends SIGTERM and immediately nulls hermesProcess), then used a hard-coded setTimeout(() => mainWindow?.reload(), 150) to trigger the renderer reload.

150 ms is an arbitrary delay that doesn't track actual process shutdown. If the Python backend takes longer to terminate (signal handler, file-lock release, cleanup code), the renderer reloads and startHermes() tries to bind the same port while the old process is still alive — resulting in an "address already in use" connection failure.

Fix

Capture the process reference before resetHermesConnection() nulls it, then await the real exit event:

• If a local backend was running: wait for it to exit, with a 5 s SIGKILL fallback in case SIGTERM is ignored.
• If no local process was alive (remote backend, already stopped): reload immediately — same behaviour as before.

const dying = hermesProcess && !hermesProcess.killed ? hermesProcess : null
resetHermesConnection()

if (dying) {
  await new Promise(resolve => {
    const timer = setTimeout(() => {
      try { dying.kill('SIGKILL') } catch {}
      resolve()
    }, 5000)
    dying.once('exit', () => {
      clearTimeout(timer)
      resolve()
    })
  })
}

mainWindow?.reload()

Test plan

• Change connection config to a remote backend while a local backend is running → renderer reloads cleanly without a "port in use" error in the logs
• Change connection config when no local backend is running → reload happens immediately (no 5 s wait)
• Kill the backend process externally mid-apply → SIGKILL fallback fires after 5 s and reload proceeds

[teknium1]

Merged via PR #39070. Your commit was cherry-picked onto current main as 6feb40e with your authorship preserved in git log. Thanks!