test: grow the integration scenario set (network aliases, depends_on default)

[lesnik512]

What

Grows the podman integration harness further with two more behavioral scenarios, per planning/changes/2026-07-13.04-grow-integration-scenarios-aliases-depends-on.md. No harness, conftest.py, pyproject.toml, or CI config changes -- each scenario reuses the existing run_pod fixture.

• Network aliases -- long-form networks: {default: {aliases: [...]}}, a distinct code path from the hostname/container_name case already covered (graph.py:_host_names's aliases-extraction branch), even though the downstream --add-host merge is already proven.
• depends_on with no condition -- defaults to service_started, the one dependency condition with zero synchronization. helper starts detached with no gate; app polls with a bounded retry loop for a marker file, rather than checking once -- the correct shape for testing a condition whose entire point is "no wait is provided, the caller bridges the gap itself." Verified locally (via direct emit_script generation, no podman) that helper emits as plain podman run -d and the retry loop's $(seq ...) survives compose2pod's own interpolation untouched, learning from the previous round's $-hijacking bug.

Both underlying claims were independently traced to source and independently re-verified during review (not merely trusted), given two of the three prior rounds in this series shipped bugs that only source-tracing or real CI caught.

Verification

just test-ci at 100% (363 passed, 10 deselected); just lint-ci and just check-planning clean. Both scenarios use busybox:1.36, already cached. Behavioral green is produced by the existing integration CI job.