test: verify the installed Cortex service survives reboot on macOS and Linux

Feature Description

A tool meant to sit in your request path all day has to come back by itself after the machine
restarts. If it does not, the user's next agent session either fails or — worse — silently
bypasses Cortex, and they lose a day of data without noticing.

This is currently one checkbox inside the macOS install issue (#944) and one inside the Linux
install issue (#945), which is how it ends up verified by neither. It also cannot be covered by
the release smoke tests (#956, #957): CI runners do not reboot, so this is deliberate manual
verification on real machines.

Proposed Solution

Verify on macOS (launchd) and Linux (systemd):

• Service auto-starts after a clean reboot, with no user action
• Service auto-starts after an unclean shutdown or hard power cycle
• macOS: comes back after logout then login, and after a reboot where the user has not yet
  logged in — state which of those launchd actually guarantees for a gui/$uid domain
• Linux: comes back under user systemd, and state what happens without enable-linger
• abctl service status reports healthy after reboot, and the proxy is listening
• The agent works after reboot without re-running enable — the settings patch and CA
  trust both survive
• Persisted sessions and metrics are intact after reboot, not truncated or corrupted (#901)
• A session interrupted mid-request by a reboot leaves readable data, not a corrupt store
• Document what a user should expect after a reboot and how to tell it worked (#960)

Additional Context

Part of the Cortex v0.9.0 release, P0. Complements #944 and #945 — those cover install, upgrade
and uninstall mechanics; this covers the machine restarting underneath a working install.

Manual verification by design. Record the platforms and OS versions actually tested.