[fix][deploy] Preserve startup failures; forward container signals

[chuandew]

Summary

• Fail the container entrypoint immediately when stop or deploy returns a nonzero status.
• Exec the existing start script so the service exit status and container signals are no longer hidden by the trailing sleep/echo.
• Add an isolated deployment regression for exit behavior and unresolved configuration variables.

The logging-variable rendering fix from #1120 is already present in the main branch. This follow-up fixes startup failure propagation and guards against unresolved configuration variables; it does not change restart policy or configuration defaults.

Verification

• bash scripts/test/test_deploy_startup.sh passes on the host and in a Rocky9 container with --ulimit nofile=1048576:1048576 --ulimit core=-1:-1.
• Before the entrypoint fix, the missing-server case returned 0 instead of 127. After the fix, missing executable, stop/deploy execution failures, normal exit, rendered configuration, and signal-exit cases all pass.
• Replacing the renderer in an isolated fixture with the exact old Jenkins [fix][monitor] Fix minor bugs #49 renderer makes the regression fail on an unresolved configuration variable.
• An existing real MDS binary rejects the unresolved log_v value but accepts the rendered configuration through its version-reporting path. This checks argument parsing, not MDS health.
• In a network-isolated container using the actual deployment/start scripts, a signal-test executable becomes PID 1 and returns its expected exit code 42 after an external SIGTERM.
• Bash syntax checks pass.

Remaining acceptance

Full Jenkins deployment, client mount, smoke/vdbench, and environment cleanup have not been rerun. Keep this PR in draft; no merge queue entry is requested.