test(docker-stack): bring up qdrant only to avoid embedding HF bake (TD-628)

[Hidden-History]

Summary

The "Integration Tests" CI job intermittently failed because tests/integration/test_docker_stack.py brought up the entire Docker Compose stack (docker compose up -d --wait with no service filter). That includes the embedding service, which bakes HuggingFace models at image-build time — and on a clean CI checkout the build hits HuggingFace's rate limiter (429 Too Many Requests), so compose up exits 1 and the tests error.

None of the tests in this file need the embedding service: they only probe Qdrant (port 26350, /healthz, /). The qdrant compose service has a pinned image: and no depends_on, so it starts standalone with no HuggingFace traffic.

Change

Scope every docker compose up in test_docker_stack.py to the qdrant service only, and adjust wait_for_services_healthy to expect Qdrant alone (it previously waited for both Qdrant and embedding to report healthy). No other files change.

This removes the embedding image build — and therefore HuggingFace — from this job's hot path, eliminating the rate-limit flake. It is also more accurate to what these tests verify (Story 1.1 Qdrant storage/port/health).

Verification

The qdrant-only bring-up path is exercised by this CI run itself (a clean checkout with no pre-existing stack). Locally, test_default_port_26350 and test_qdrant_service_accessible pass.

Addresses the recurring Integration Tests flake (also intermittently red on main). The separate Ubuntu install.sh HF-429 case is tracked independently and not in scope here.