chore(deploy): remove local images after push to bound runner disk

[jaeyunha]

Why

The shared Mac mini self-hosted runner (forever / opensend / exponential all share one Docker daemon) keeps filling its disk. scripts/deploy-ecs.sh builds <ecr>/exponential-{api,web}:<sha> locally then pushes, but never removes the local copies. docker image prune only reaps dangling images, so these tagged per-commit builds leak disk on every deploy.

Fix

A best-effort EXIT trap that deletes this deploy's own api + web image tags after the run. Runs on EXIT (fires even if a later step fails); || true so it can never fail the deploy. The build/push/ECS logic is untouched.

Mirrors the same fix already merged in forever-agent#420 (verified working on a live deploy).

Validation

• bash -n scripts/deploy-ecs.sh ✅
• Trap references only ECR_REGISTRY/APP_NAME/IMAGE_TAG, all defined before it.

Heads-up

This path (scripts/deploy-ecs.sh) is in the deploy workflow's paths: filter, so merging will trigger a deploy — expected and safe.

🤖 Generated with Claude Code

[jaeyunha]

Controller disposition for current head 2830707: human-gated; do not merge in this ordinary staging tick.

Evidence:

• PR chore(deploy): remove local images after push to bound runner disk #712 targets main directly and changes scripts/deploy-ecs.sh; the PR body notes that merging this path triggers a deploy.
• Diff is one deploy-script change and cheap gates passed: git diff --check origin/main...pull/712/head and bash -n scripts/deploy-ecs.sh.
• No GitHub checks are reported.

Blocker:

• This is production deploy behavior on main, not an ordinary implementation PR for staging; it needs explicit owner routing.

[Jaeyun Ha] Should this deploy-runner disk cleanup be retargeted/rebuilt through the staging release flow, or is it approved as a direct main hotfix/deploy change?