Fix UI restart hanging before response flush

[ruizanthony]

Summary

• Make /api/restart return immediately and schedule the actual reload in a delayed daemon thread.
• Prevent duplicate reload requests while one is already in progress.
• Use in-process restart via os.execv() for normal UI restarts, so the response can flush cleanly.
• Detect pending native self-updates via /exe/a0-self-update.yaml and hand control back to the native supervisor instead of re-execing run_ui.py.
• When running under systemd, request systemctl restart --no-block agent-zero.service so the service wrapper can consume the self-update trigger.
• Allow HEAD /api/health for readiness probes.

Why

The Web UI restart flow could hang because /api/restart called process.reload() before the HTTP response was reliably flushed. In Docker/LXC this could terminate or interrupt the request, leaving the UI polling/waiting indefinitely.

A follow-up issue was found with self-update: if /exe/a0-self-update.yaml was pending, the new UI restart path could directly re-exec run_ui.py and bypass the native wrapper/self-update manager that applies the update. In systemd deployments, the active UI process can also be a child process rather than the service MainPID, so simply exiting the UI process may not restart the whole service. The restart helper now detects this pending self-update case and asks systemd to restart the service before exiting, with a supervisor-exit fallback when systemd is unavailable.

Validation

• Rebased on current origin/main / v1.14.
• python -m py_compile api/restart.py api/health.py helpers/process.py tests/test_restart_api.py tests/test_process_reload.py
• python -m pytest -q tests/test_restart_api.py tests/test_process_reload.py → 12 passed