Dev control-plane rollout fails ELB health

The Jaah Deploy to dev has not landed since about 2026-10-29 01:20Z. Dev is still on task definition jaah-dev-cp:1406. - Run 40800827524 (develop 41e1bb737): `apply (dev)` failed. The new CP tasks went target-healthy in 50–110s, then failed ELB health checks with 502 on both target groups (jaah-dev-cp on port 3050, and jaah-dev-cp-public on port 3051). ECS replaced them 3 times and then rolled back to :1406. Log: `jaah-forge ci log 123212544341`. - Run 40799965317 (develop 337c84445) failed earlier on a transient error: "could not connect to registry.terraform.io". Suspects, since the last deploy that served: the API-side community-edition cuts. - 34c911dc3 "feat(ce): the community edition's api answers only its own machine (Task 11163)". The ALB health check comes from another address. - 166386d24 / fbcb97618 "cut aws/ and bedrock/ from the community edition tree (Task 11151)". - f40966365, b6656ea4b and dca9486f9 (CE tree scrubs). Find which change makes a dev task stop answering the ALB health check after about 2 minutes, fix it, and get Jaah Deploy (dev) green. Everything merged since 41e1bb7 is blocked from dev until then. That includes the task-templates epic (11051), whose AC-20 needs the dev deploy. Out of scope: changes to the templates feature itself. Done means: - A Jaah Deploy (dev, apply) run on develop completes `success`. - `https://app.dev.jaah.ai/templates/manage` renders the manage list as the operator.

The fix had already shipped. The dev rollback came from migration 202610290030's slow cascading DELETE, which outlasted the ELB health checks; b9e4f1cb0 (Task 11184) fixed it. Dev deploys have been green since 02:44Z, the current develop head is live, and /templates/manage renders for the operator. Memory: none applied.