feat: add Docker image layer caching between CI runs

[chorrell]

Summary

Caches Docker image layers between CI runs using the GitHub Actions cache backend, so repeated pushes build significantly faster.

What changed

.github/workflows/ruby.yml

• Adds docker/setup-buildx-action to enable the GHA cache backend
• Adds two docker/build-push-action steps that build node-test:22 and node-test:24 with cache-from/cache-to: type=gha before running tests
• Removes the now-redundant DOCKER_BUILDKIT: 1 env var (Buildx enables BuildKit by default; Rakefile also sets it)
• Suppresses a low-confidence zizmor cache-poisoning finding — both Docker actions are pinned to exact commit SHAs

spec/spec_helper.rb

• create_image now tries Docker::Image.get("node-test:#{version}") first; falls back to build_from_dir if the pre-built image isn't present (e.g. local development)

How caching works

• First run: layers are built from scratch and written to the GHA cache under scopes node22 / node24
• Subsequent runs: cache-from: type=gha restores the cached layers — only changed layers are rebuilt
• Both Ruby matrix jobs (3.4 and 4.0) share the same cache scopes, so the second job benefits from the first job's cache write on the same commit

Test plan

• First CI run builds from scratch and writes cache
• Second run shows cache hits in the "Build Node.js XX image" steps
• Tests still pass on both Ruby 3.4 and 4.0

🤖 Generated with Claude Code