chore(infra): move staging from localstack to floci

DO NOT MERGE before the Floci cutover window. runtime.env points deploys at the path-style ECR on localhost:4566, which does not exist until Floci replaces LocalStack on holapc. Merging earlier makes the develop pipeline's deploy:staging fail. The first deploy after cutover must be this merge commit.

Why: the shared LocalStack on holapc runs on a Pro trial that ends 2026-10-24 03:19 UTC; ECR is Pro-only. Floci (MIT) replaces it for moanapay and academy.

• runtime.env: path-style ECR_REGISTRY and repositories (localhost:4566/000000000000/us-east-1). ci-templates v1 sources runtime.env first, so no job override is needed; docker login with the path value was verified against the PoC Floci ("Login Succeeded").
• deploy.sh: docker login uses the registry host only.
• edit-env.sh: docker exec … awslocal replaced by an amazon/aws-cli container on the network; --rollback-env relies on AWSPREVIOUS, verified on Floci.
• Terraform/BE/compose: LocalStack comments updated; hostname localstack and network ahad-localstack_default kept (Floci takes them over).
• No ignore_changes added: the module is shared with prod. Floci leaves 5 harmless non-converging in-place updates in the staging plan (documented in elasticache.tf).
• Known gap: Redis AUTH without TLS on staging (Floci has no Redis TLS). app-env at cutover: DB_PORT=7001, REDIS_PORT=6379, REDIS_PASSWORD, REDIS_TLS=false.

Review gate: Opus, PASS in round 2.

🤖 Generated with Claude Code