feat: adopt mitodl/ol-python-base:3.12 and add BuildKit cache mounts

[blarghmatey]

What are the relevant tickets?

N/A

Description (What does it do?)

⚠️ Draft — blocked on mitodl/ol-infrastructure#4750 merging and mitodl/ol-python-base:3.12 being published to DockerHub.

Adopts the new shared Python base image and adds BuildKit cache mounts to speed up cold-build and uv.lock-bump rebuilds in the Tilt/K3d local-dev environment.

Changes:

• FROM mitodl/ol-python-base:3.12 replaces the four duplicated substrate stages (apt core, uv binary, mitodl user, /opt/venv env vars). The shared base is defined in https://github.com/mitodl/ol-infrastructure/tree/main/dockerfiles/ol-python-base.
• App-specific apt (libcairo2-dev, poppler-utils, default-jre, chromium, chromium-driver) kept inline with --mount=type=cache for apt so downloaded packages are not baked into the image layer.
• --mount=type=cache,target=/opt/uv-cache,uid=1000,gid=1000 on uv sync so uv.lock-bump rebuilds reuse the download cache across Tilt builds rather than re-downloading all wheels.
• apt.txt trimmed to extras only; the common-core packages it previously listed are now in the base image.
• Multi-stage structure preserved (base → deps → final).

Image size: 4.36 GB → 3.71 GB (−650 MB), which cuts the ~119s registry push proportionally.

How can this be tested?

After mitodl/ol-infrastructure#4750 merges and the Concourse pipeline publishes the base image:

# From the mit-learn repo root
DOCKER_BUILDKIT=1 docker build -t mitodl/mit-learn-app:test .
docker run --rm mitodl/mit-learn-app:test python --version
# Confirm uv sync cache reuse: touch uv.lock and rebuild — second build should
# skip all wheel downloads.

In the local-dev Tilt environment: tilt up mit-learn, confirm live_update still syncs Python source and triggers granian reload.

Additional Context

Part of a multi-repo series to eliminate duplicated Dockerfile substrate and speed up Tilt/K3d cold builds. Baseline: cold tilt up for mit-learn spent ~163s building and ~119s pushing its 4.36 GB image to the local k3d registry at localhost:5001. The shared base image (~590 MB) plus per-app layer slimming reduces both dimensions.

Merge order: ol-infrastructure PR first → Concourse pipeline registered → this PR can be merged.

[github-actions]

OpenAPI Changes

No changes detected

View full changelog

Unexpected changes? Ensure your branch is up-to-date with main (consider rebasing).