What changed and why

Calling systemctl (--user) restart makes a synchronous D-Bus call that blocks indefinitely when the D-Bus session bus loses its connection during the restart handoff. This reproduces reliably on Ubuntu 24.04 ARM64 with a Linuxbrew-installed Hermes user unit (issue #29421).

Per the reporter's diagnostics on 2026-05-21: both hermes gateway restart and the raw systemctl --user restart hermes-gateway hang, while systemctl --user --no-block restart hermes-gateway returns immediately (exit 0), pointing squarely at the D-Bus job-wait.

Added --no-block to every systemctl restart invocation inside systemd_restart() (three call sites: post-graceful-drain, forced-restart fallback, and cold-start). The existing _wait_for_systemd_service_restart() poll loop already handles waiting for the new PID and printing progress messages, so operator-visible behaviour is unchanged.

How to test

1. Install Hermes as a systemd user unit on Ubuntu 24.04 ARM64 (or any host that exhibits the D-Bus session bus drop during restart).
2. Run hermes gateway restart — it should complete and print ✓ User service restarted (PID N) instead of hanging.
3. Alternatively: strace -e trace=sendmsg,recvmsg hermes gateway restart — confirm that the systemctl child exits quickly rather than blocking on D-Bus receives.

What platforms tested on

• macOS 14 (unit tests only; systemd tests skipped due to platform)
• CI (Linux): existing test suite passes; 5 pre-existing macOS-only systemd preflight failures are unrelated and unchanged from main

Fixes #29421