v0.4.9: fix the two defects the 3-day trial found

[deva0x]

Two defects found by a three-day trial run on a real Mac and a real Ubuntu box — neither was findable by review or CI.

1. Upgrading to v0.4.8 set off a false alarm on every listener.
v0.4.8 started storing *:5000 where older snapshots held 5000. The diff compared those as opaque strings, so the first run after upgrading reported 11 "changed listener" findings on the Mac and 3 on the Linux box — all ORANGE, all above the --notify threshold, and every one false: the ports were identical. The v0.4.8 changelog predicted this churn and waved it through as harmless. It was not.

The listening diff now compares what both sides actually express: when either side predates the change it compares port sets, so a pure representation change yields nothing. Deliberately not a blanket exemption —

• a genuinely new port arriving alongside the migration is still reported at ORANGE;
• once both snapshots carry addresses, 127.0.0.1:5000 → *:5000 is a real exposure change and stays a finding.

Both guarantees have tests, and both tests were mutation-checked (removing the fix, and over-applying it, each fail).

2. The Linux installer promised a daily job it could not keep.
systemctl --user enable --now succeeding proves only that a user manager exists at that moment — and it exists because you are logged in. With lingering off, systemd tears that manager down at your last logout and the timer goes with it, so the installer printed a confident "enabled systemd user timer" for a job that then goes dormant on exactly the machine that needs it most: a server nobody logs into. (Persistent=true defers the run to the next login rather than losing it.)

The installer now checks loginctl show-user … -p Linger after a successful enable and prints the one-line fix. Verified on a real Ubuntu 24.04 box in both directions: the note appears with Linger=no, and is absent after enable-linger.

Suite 689 → 692.

🤖 Generated with Claude Code