Stop deploys from resetting the ECS task count set by the autoscaler

[chelseakr-exygy]

Description

Every bin/deploy set the ECS service back to FARGATE_DESIRED_TASK_COUNT, draining tasks the autoscaler had added. The service now ignores later changes to desired_count, so the value only sets the starting count.

Behavior change: setting FARGATE_DESIRED_TASK_COUNT=0 and redeploying no longer shuts the server down. Use aws ecs update-service --desired-count <n>. The variable description is updated, and civiform-deploy's example config needs a follow-up.

This is a technical change to the deployment infrastructure only. There's no change to CiviForm itself and nothing for QA to test.

Code written with AI assistance (Claude), PR body written by human!

Checklist

General

• Added the correct label
• Assigned to a specific person or civiform/deployment-system
• Created tests which fail without the change (if possible)
• Performed manual testing (at a minimum run bin/setup without your changes and then bin/deploy with your changes to ensure your changes don't break existing deployments)
• Extended the README / documentation, if necessary

Instructions for manual testing

Technical check for the deployment reviewer, not a QA task. Not yet run in AWS.

1. bin/setup and bin/deploy on main with FARGATE_DESIRED_TASK_COUNT=1.
2. Scale the service to 3: aws ecs update-service --cluster <cluster> --service <service> --desired-count 3.
3. bin/deploy from this branch. The plan should show no desired_count change, and the service should stay at 3 tasks through the rollout.
4. Set FARGATE_DESIRED_TASK_COUNT=2 and deploy again. No change should be proposed.
5. Change a feature flag and deploy. The task count should not drop during the rollout.

Issue(s) this completes

Fixes civiform/civiform#14017

[chelseakr-exygy]

forgot to perform manual testing, stand by