feat: integration test pipeline with containerized services

[charlie83Gs]

Goal

Run integration tests in CI that exercise cross-service behavior by booting services as container sidecars, similar to how we run Postgres and Redis today.

Background

Worker packages have circular imports when installed in a single venv (see #1). Integration tests that cross service boundaries need isolated runtime environments. The solution is to run each service in its own container during integration test jobs.

Proposed approach

1. No-op base pipeline

Create a base Docker Compose or GitHub Actions service configuration that boots:

• All infrastructure (Postgres graph-db, Postgres write-db, Redis, Hatchet)
• Worker services as containers (using the images from ghcr.io/openktree/knowledge-tree/<service>)
• API service

Each service runs in its own container with its own Python environment, eliminating circular import issues.

2. Integration test job in CI

Add a new job to test.yml (or a separate integration-test.yml):

integration-test:
  name: Integration Tests
  runs-on: ubuntu-latest
  services:
    postgres: ...
    postgres-write: ...
    redis: ...
    # Boot workers as containers
    worker-orchestrator:
      image: ghcr.io/openktree/knowledge-tree/worker-orchestrator:latest
    worker-nodes:
      image: ghcr.io/openktree/knowledge-tree/worker-nodes:latest
    api:
      image: ghcr.io/openktree/knowledge-tree/api:latest
  steps:
    - uses: actions/checkout@v4
    - run: uv run pytest tests/integration/ -x

3. Test structure

tests/
  integration/
    test_exploration_e2e.py      # Full exploration flow
    test_node_pipeline_e2e.py    # Node creation pipeline
    test_sync_e2e.py             # Write-db → graph-db sync
    conftest.py                  # Fixtures pointing to container URLs

Acceptance criteria

• No-op base pipeline config that boots all services as containers
• Integration test job in CI that runs after Docker images are built
• At least one end-to-end test exercising the exploration → node creation flow
• Integration tests only run on main (not PRs) to keep PR feedback fast