fix: retry Linux background launcher after early exit

[jamilahmadzai]

Summary

• retry non-launchd POSIX launcher starts when the child exits before readiness
• keep retries bounded and inside the existing readiness deadline
• add a Linux regression that proves the second start is used after an early child exit

Root cause

The macOS path delegates non-zero agent exits to launchd KeepAlive, but the Linux/background path started the child once. If that child exited before the health endpoint became ready, the launcher observed the dead PID, stopped waiting, and returned a not-ready error without giving the agent another chance to recover.

The retry stays limited by MONK_AGENT_BACKGROUND_RESTARTS with a small default and the existing readiness deadline, so persistent failures still terminate cleanly.

Validation

• ./tests/start-monk-agent-background-retry.sh
• ./tests/start-monk-agent-readiness-timeout.sh
• ./tests/start-monk-agent-fastpath.sh
• sh -n scripts/start-monk-agent.sh plugins/monk/scripts/start-monk-agent.sh .antigravity-plugin/scripts/start-monk-agent.sh tests/start-monk-agent-background-retry.sh tests/start-monk-agent-readiness-timeout.sh tests/start-monk-agent-fastpath.sh
• generated launcher copy parity with cmp
• Ruby YAML parse for .github/workflows/install-e2e.yml
• git diff --check

Fixes #180.