[Artifact cache] Integrate CI planning and verified artifact reuse #9533

Goal

Replace or augment baseline-run compatibility decisions with fingerprint-based artifact selection.

Work

First integrate dry-run planning:

• Report candidates, misses and skipped outputs.
• Report predicted build decisions.
• Preserve existing build execution.
• Compare predictions with verified artifact availability.

After validating predictions:

• Connect verified restoration to CI consumer builds.
• Augment build_tools/github_actions/baseline_runs.py.
• Modify the existing prebuilt-copy workflow path.
• Permit compatible artifacts to come from different producing run IDs.
• Require complete expected stage outputs before skipping a stage.
• Preserve normal-build fallback for unavailable or invalid cache outputs.
• Keep selected tests independent of cache availability.

Treat building only missing subprojects within a partially cached stage as a separate optimization unless its dependency handling is proven.

Acceptance criteria

• Dry-run predictions match verified availability.
• Metadata candidates alone cannot skip jobs.
• All required outputs are restored before a stage is skipped.
• Mixed-hit and unavailable-cache cases retain correct build behavior.
• Artifacts from different runs are combined only when compatible with current inputs.
• Selected tests still execute.
• Reports distinguish predicted reuse from actual reuse.

Dependencies

Task 9 and the completed local consumer-build integration.