feat: notification_all_screens config + event-driven overlay dismissal

[ctsstc]

Summary

Add a new config option to display notifications on all screens or only the active screen.

Reason

Themes only show on the active screen.
The default theme/no theme selected shows the notification on all screens.
Currently there's no parity across the system, this brings it all together and lets the user choose behavior.

Changes

• Add new config option: notification_all_screens
  • Add new config to migration scripts
    • Keep previous functionality
      • If using the default theme default to: true
      • If using one of the themes default to: false
• Add IPC via: NSDistributedNotificationCenter for notification dismissal

[vercel]

@ctsstc is attempting to deploy a commit to the Gary Sheng's projects Team on Vercel.

A member of the Team first needs to authorize it.

[garysheng]

Thanks for this — the notification_all_screens config and event-driven dismissal via NSDistributedNotificationCenter are solid additions, and the new mac-overlay.bats tests are thorough.

One issue blocking merge: test 537 in peon.bats now fails because the PR appends all_screens as a new last argument to the overlay call. The existing test uses awk '{print $(NF-2)}' to extract ide_pid, expecting notif_position to be the last field — but with the new arg, NF is now all_screens, shifting ide_pid to NF-3.

Fix needed in tests/peon.bats around line 2879:

# Update the comment and offset to account for the new all_screens arg at the end
# Fields from end: ... ide_pid session_tty notif_position notify_type all_screens
# ide_pid is NF-4 (or adjust to use a fixed field index)
ide_pid=$(tail -1 "$TEST_DIR/overlay.log" | awk '{print $(NF-4)}')

The Windows respects custom debug_retention_days from config failure is pre-existing on main and not introduced by this PR.

Once the peon.bats fix lands, this is ready to merge.

[garysheng]

The Windows CI failure ("respects custom debug_retention_days from config") was a pre-existing flaky test caused by hardcoded dates — not related to your changes. It was just fixed in #420 which merged moments ago.

Could you rebase your branch onto the latest main? That should bring in the fix and get Windows CI to green.

The feature logic itself looks solid — looking forward to getting this merged once CI is clean.

[ctsstc]

@garysheng looks good to go now