Reject empty coverage evidence

[Anionix]

Summary

• select coverage by the imported format_bench package identity instead of the unmatched repository src path
• make the report step fail with status 2 when total coverage is below 1%, preventing a 0% empty artifact from passing
• add a cheap workflow-contract regression test

Trigger and impact

Main run https://github.com/Anionix/data-format-lab/actions/runs/30079480822 passed 598 tests but emitted CoverageWarning: No data was collected and TOTAL ... 0%. Its uploaded artifact therefore could not establish #303 baseline evidence.

Primary contracts

• coverage.py source selection accepts importable package/module names
• coverage report --fail-under=MIN exits 2 below MIN
• coverage.py 7.10.7 and uv one-off dependency links are pinned in workflow comments
• coverage is descriptive code-quality evidence and never advances benchmark lifecycle or ranking eligibility
• NO_LEAN: workflow/test-only change; no Lean definitions or proofs

Verification

Exact head: f946e0e318451093a1a0d7c3e00e0427c72c3ce7

• focused workflow regression: 1 passed
• Ruff: clean
• workflow YAML: parsed with coverage job present
• fixed-SHA spec review: no findings
• fixed-SHA standards/code-smell review: no findings

Post-merge: dispatch the workflow on the merged main SHA; require nonzero TOTAL and a non-expired exact-SHA artifact before closing #303.

Fixes #508
Refs #303
Refs #506