feat: add shell -r/--restart for race-free shell restarts

[dadams-AU]

Implements option 2 from #139, as discussed there.

What it does

caelestia shell -r (composes with -d and --log-rules):

1. If an instance is running, asks it to quit (qs -c caelestia kill).
2. Polls qs -c caelestia list -j until the instance actually disappears — teardown is not instant, and is slowest while a session lock is up, which is exactly when a reliable restart matters most.
3. If the instance is still there after 5s (hung shell), escalates to SIGKILL on the recorded PIDs (with a warning), waits again briefly, then proceeds.
4. Starts the shell as usual.

With no instance running it behaves like a plain start, so it's safe as the default "get me a shell" keybind.

Why not just propagate the exit status (option 1)

While implementing this I found that qs -n itself exits 0 on the duplicate no-op:

$ qs -c caelestia -n -d; echo $?
An instance of this configuration is already running.
0

so propagating subprocess.run's return code from shell -d wouldn't have helped — polling qs list is the only reliable way to detect the case. That makes --restart the fix rather than a convenience.

Testing

On a live session (Hyprland, quickshell-git):

• shell -r -d with a running instance: old instance killed and fully gone, new instance up and IPC-responsive, exactly one instance in qs list afterwards.
• The race from the issue: qs -c caelestia kill; sleep .1; caelestia shell -r -d (restart requested mid-teardown) — waits out the dying instance and launches cleanly, where shell -d previously no-op'd and left zero shells.
• Regression: plain shell -d with a running instance is unchanged (message + exit 0, no second instance).

Also adds the flag to the fish completions.

🤖 Generated with Claude Code

[dark3txr]

read contributing, NO AI CODE