fix(gateway): remove -k flag from kickstart in launchd_restart to prevent double-spawn

[liuhao1024]

What does this PR do?

Fixes launchd_restart() to use plain kickstart (without -k flag), preventing a double-spawn race condition on macOS 26+ when KeepAlive=true is set in the LaunchAgent plist.

Related Issue

Fixes #42446

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Root Cause

In launchd_restart() (hermes_cli/gateway.py), the code already performs graceful termination:

1. terminate_pid() sends SIGTERM to the gateway
2. _wait_for_gateway_exit() waits for the process to exit

Then kickstart -k is called, but -k means "terminate if running, then start." This races with KeepAlive=true:

1. SIGTERM sent → gateway begins shutdown
2. kickstart -k detects running process → terminates again
3. KeepAlive=true detects termination → respawns
4. kickstart also starts a new instance
5. Result: two concurrent gateway processes

Changes Made

• hermes_cli/gateway.py: Removed -k flag from kickstart in launchd_restart() (line 3540), since the preceding code already handles termination. This matches the pattern used by all other kickstart calls in the codebase (launchd_start paths and the fallback path in the same function).
• tests/cli/test_launchd_restart_kickstart.py: Added 4 regression tests verifying the kickstart command does not include -k, that termination happens before kickstart, and that self-restart (SIGUSR1) skips kickstart entirely.

Testing

$ python -m pytest tests/cli/test_launchd_restart_kickstart.py -v
tests/cli/test_launchd_restart_kickstart.py::test_kickstart_uses_no_kill_flag PASSED
tests/cli/test_launchd_restart_kickstart.py::test_kickstart_no_kill_flag_when_pid_absent PASSED
tests/cli/test_launchd_restart_kickstart.py::test_terminate_called_before_kickstart PASSED
tests/cli/test_launchd_restart_kickstart.py::test_self_restart_skips_kickstart PASSED

4 passed in 0.13s

Checklist

• I have read the contribution guidelines
• My code follows the project's coding standards
• I have added tests that prove my fix is effective
• All new and existing tests passed

[teknium1]

Thanks for the focused macOS report and regression coverage.

Problems

• launchd_restart() deliberately retains kickstart -k for the failed-drain branch: current main logs a forced restart when _wait_for_gateway_exit() returns false at hermes_cli/gateway.py:4347-4351, then runs kickstart -k at hermes_cli/gateway.py:4352. Making the command plain unconditionally removes that fallback.
• The added tests do not exercise _wait_for_gateway_exit(...)=False; test_terminate_called_before_kickstart verifies calls occurred, not their ordering or the timeout behavior.

Suggested changes

• Split the command choice by the drain result: preserve a hard restart after timeout, and cover both branches in tests/hermes_cli/test_gateway_service.py's existing ordered-call tests.

Automated hermes-sweeper review.

[teknium1]

Please keep a forced-restart branch when _wait_for_gateway_exit() returns False. Current main explicitly logs that the drain timed out and then uses kickstart -k to force the restart; making this plain unconditionally removes that fallback.