[app] Graceful shutdown consistency (grace periods and preStop alignment)

[henzigo]

Context

Grace periods and preStop hooks are inconsistent across workloads:

Container	Grace	preStop	Issue
php-fpm	120 s	sleep 10; kill -QUIT 1	grace longer than needed (10 s drain + 60 s request_terminate_timeout)
nginx	120 s	sleep 5 && nginx -s quit	drains 5 s while php-fpm drains 10 s — nginx stops accepting before FPM
storefront	60 s	sleep 10	Node exits in seconds; ~47 s wasted on every rollout
redis	30 s (default)	none	no explicit shutdown; 2 GB in-memory store

Proposal

Align (all as component values, these become the defaults):

• php-fpm grace 90 (10 drain + 60 worker timeout + buffer)
• nginx preStop sleep 10 && nginx -s quit (same LB-drain window as php-fpm)
• storefront grace 30
• redis grace 60 + preStop redis-cli SHUTDOWN NOSAVE (ephemeral cache — NOSAVE)

Cron and RabbitMQ are handled in #25 and #6 respectively — keep this issue to the four above.

Tasks

• Defaults update + values plumbing where missing (redis preStop/grace)
• Unit tests, golden snapshots, deviations entry
• Rollout observation on a dev cluster (no 5xx during rollout, faster drains)