Summary

First slice of the ECS CD-pipeline work (Phase 0+1 of the plan). Independent value on its own — fixes two real bugs found while auditing deploy-to-ecs.yml.

• api/views/health.py: health_check returned HTTP 200 unconditionally (JsonResponse(data), no status=), even when overall_status was False. The ECS task definition's container healthcheck (curl -f http://localhost:8000/health/) and any future smoke gate hitting this endpoint would both false-green a container with a dead DB/ES/Redis/telemetry connection. Now returns 503 when unhealthy.
• Added a git_sha field to the same response, sourced from a new GIT_COMMIT_SHA env var (plumbed via ARG/ENV in the Dockerfile, same pattern already used for other build-time config). This is the primitive a later smoke gate needs to verify "the code that's live is the code that was just pushed."
• docker-entrypoint.sh: dropped python manage.py makemigrations --noinput, which ran on every container boot. Generating migration files at deploy time instead of using committed ones is unsafe under Fargate's rolling deployment (briefly 2 tasks live) and means the schema landing in prod was never reviewed. migrate --noinput stays in the entrypoint for now — moving it to an explicit one-off pipeline step is a separate follow-up PR.
• Also deleted a local, never-committed deploy-backend.yml draft (EC2 + GHCR + docker compose) that had no matching secrets/infra behind it — confirmed the real deploy target is ECS Fargate (CloudFormation-provisioned RDS/ElastiCache/managed-ES, already-configured development environment secrets), not EC2.

Test plan

• tests/test_health.py (new, 5 tests): 200 when all deps healthy, 503 when ES down, 503 when Redis down, git_sha defaults to "unknown", git_sha reflects GIT_COMMIT_SHA
• Full suite run locally (pytest tests/): 100 passed vs. the pre-existing baseline's 95 passed — no new failures. The 22 failed / 7 errored tests in charts/test_dataset_schema/test_geography_hierarchy are pre-existing on dev (confirmed by stashing this branch's changes and re-running against unmodified dev — identical failure set)
• bash -n + shellcheck clean on docker-entrypoint.sh
• python -m py_compile clean on health.py

Next in this series

Moving migrate out of the entrypoint into an explicit ECS one-off task, fixing deploy-to-ecs.yml's if: always() and fragile infra-change condition, wiring a smoke-test gate, and adding rollback-on-failure — tracked as follow-up PRs.