Optimize production memory allocation

[ewlarson]

Summary

• double the production Celery worker memory ceiling from 2 GiB to 4 GiB
• raise the production Redis data ceiling from 12 GB to 16 GB and bound the container at 18 GiB
• bound Elasticsearch at 8 GiB while retaining its configured 4 GiB JVM heap
• add a regression test for the production memory profile

Rationale

This aligns the production service ceilings with the expanded host capacity while retaining headroom for PostgreSQL, filesystem cache, the proxy, and host monitoring services. Web and cron allocations remain unchanged.

Deployment note

The worker ceiling takes effect when the application role is redeployed. Redis and Elasticsearch settings take effect when those accessory containers are upgraded or recreated; merging this PR alone does not change running production services.

Testing

• pytest backend/tests/test_kamal_deploy_config.py -q (14 passed)
• make lint-check
• merged Kamal production configuration renders successfully with the intended memory settings