Add multi-arch pre-built images for faster builds

[Copilot]

Lenny's build process requires 10-20 minutes to compile all services from scratch. This PR pre-builds multi-architecture (amd64/arm64) images for static services, reducing build time to 2-5 minutes.

Changes

GitHub Actions workflow (.github/workflows/build-images.yml)

• Builds lenny-api for linux/amd64 and linux/arm64
• Triggers on changes to docker/, compose.yaml, lenny/, or requirements.txt
• Pushes to GitHub Container Registry with caching
• Uses GITHUB_TOKEN (no additional secrets required)

Updated compose.yaml

• API service references ghcr.io/archivelabs/lenny/lenny-api:latest with local build fallback
• Reader service remains local-only build (requires dynamic NEXT_PUBLIC_MANIFEST_ALLOWED_DOMAINS at build time)
• Added comments clarifying pre-built vs local-build services

Makefile

• Added make pull target to update pre-built images

Documentation

• DOCKER_IMAGES.md: Build system architecture, repository setup instructions, manual build procedures
• README.md: Notes on faster setup and multi-arch support

Architecture

Service	Strategy	Reason
API	Pre-built + local	Static config, no build-time env vars
Reader	Local only	Requires dynamic NEXT_PUBLIC_MANIFEST_ALLOWED_DOMAINS
DB/S3	Pre-built	Official upstream images
Readium	Pre-built	Official upstream image

Repository Setup

Post-merge actions:

1. Set workflow permissions to "Read and write" in Settings > Actions > General
2. Make package public at github.com/orgs/ArchiveLabs/packages/container/lenny%2Flenny-api

First workflow run on main will populate the registry. Users on subsequent clones automatically benefit from pre-built images.

Original prompt

Builds take a long time for Lenny via the Makefile + compose.

We'd like to make it faster by prebuilding images for Lenny that are multi-architecture (i.e. linux/amd64,linux/arm64).

One challenge is that, when we're building, Lenny has an option to use a custom domain or cloudflared tunnel, which ends up being a env variable that needs to be passed to lenny-reader to be built.

In an ideal world, what would happen is that the minio, db, lenny-api, and readium images would all be somehow prepared and built and the only thing the client would have to do is run the Makefile and have the lenny-readium image + container get built via the Makefile.

Generally we won't need to rebuild images unless there's a change to the domain/tunnel ENV or a change to the compose.yml or something in the docker/ directory that changes how the project is built or run.

We may need to add a new key to github so we can upload the images to github's docker repository. Can you prototype this PR

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.