test: strengthen local CLI coverage

[TrevorBasinger]

What this changes

This PR raises confidence in the real roar CLI by shifting more of the suite toward local product-path coverage instead of thin mocked wrappers.

It adds subprocess coverage across the main CLI surfaces, brings the stronger local integration layer into the default gate, and trims duplicated wrapper and DAG happy-path tests so the suite stays focused on real command behavior.

It also adds automatic local labels for detected dataset composite artifacts. When roar detects that a run output is a dataset-like composite, it now attaches system-managed dataset labels to that artifact automatically. This starts moving dataset identity toward the label system without removing the existing metadata path yet.

Why

The main goal is to make the default local test signal better reflect what users actually run. That means preferring end-to-end CLI paths where possible, keeping mocks for narrow assumptions only, and reducing duplicated scenarios that slow the suite down without adding much confidence.

The dataset label change is part of the same direction: detected datasets should become queryable and discoverable through labels automatically, rather than relying only on artifact metadata.

Notes

Existing non-dataset labels on an artifact are preserved. The auto-managed dataset label subtree is only updated when the detected dataset values change.

Testing

• ruff check .
• mypy roar
• pytest -m "not live_glaas and not ebpf"

[christophergeyer]

Cool - I'll try to run a smoke test on it.
Might there be conflicts with the backend speedup PR which batches db inserts?

[TrevorBasinger]

Cool - I'll try to run a smoke test on it. Might there be conflicts with the backend speedup PR which batches db inserts?

I just approved and merged that PR for you. Looks likes it didn't create any conflicts for this PR.