fix: wait for POSIX companion shutdown before restart

[jamilahmadzai]

Summary

• wait up to 10 seconds for the managed POSIX companion to stop before launching its replacement
• fail safely if the old process does not stop, instead of starting a competing process
• keep the root, packaged Monk, and Antigravity launcher copies byte-identical
• add a deterministic restart-handoff regression to Linux and macOS CI

Root cause

The POSIX background launcher sent SIGTERM to the recorded process and immediately started the replacement. If the old companion still owned the port, the replacement could defer to the apparently healthy old process and exit. The launcher's first readiness check could then succeed against the process being stopped, so SessionStart returned success shortly before both companions were gone.

Waiting for the managed process to exit closes that handoff window. The bounded timeout also prevents the launcher from hanging indefinitely or creating a competing replacement when shutdown fails.

Validation

• ./tests/start-monk-agent-restart-handoff.sh (five consecutive passes)
• sh -n on all three launcher copies and the new regression
• byte-for-byte comparison of all three launcher copies
• workflow YAML parse
• git diff --check

Fixes #154.