docs: document root cause of sandbox crash loop

feat: enable HUNGRY_METRICS=1 in AGENTS.md commands

fix: add CloudFront CDN + headless_shell binary to delivery policy

fix: add ssh -n to prevent stdin consumption in curl|bash

fix: wait 60s after onboard for sandbox to stabilize from presets

fix: clean duplicate SSH config entries before adding new one

fix: correct set +eo pipefail syntax (must be two separate commands)

fix: apply delivery service + playwright policies DURING onboard

fix: disable set -e before sandbox stabilization wait

fix: wait 30s after policy change, disable set -e for deploy

fix: show actual errors instead of failing silently

fix: wait for sandbox to stabilize after policy change

fix: always run nemoclaw onboard with RECREATE_SANDBOX=1

fix: show full nemoclaw onboard output during install

fix: force-update repos on reinstall (git reset --hard origin/main)