Workspace tooling tests

Verification for the coverage tooling (Workspace test-coverage mea... (FEAT_0120)). Unlike crate tests, these are procedural verifications of developer tooling: each describes the check, how it was executed, and the observed evidence. The CI-facing cases (Full coverage run produces ... (TEST_0975), CI measures and publishes o... (TEST_0979)) re-execute on every code-changing PR via .github/workflows/ci-coverage.yml; the one-shot cases were executed at introduction (2026-07-03) and are cheap to repeat locally.

Test Case: Full coverage run produces the three reports TEST_0975
status: implemented
verifies: REQ_0991, REQ_0992, REQ_0993, REQ_0995
links incoming: REQ_0991, REQ_0992, REQ_0993, REQ_0995

scripts/coverage.sh runs end-to-end: one cargo llvm-cov invocation with --workspace --all-features and -- --test-threads=1, followed by lcov, HTML, and terminal-summary reports from the same profile data. Verified by the baseline run (2026-07-03: 87.7% line coverage, all tests passing, all three reports present under target/llvm-cov/) and re-executed by every ci-coverage.yml run — first CI evidence: actions run 28669729394 (PR #182), which uploaded the 2.2 MB coverage-report artifact.
Test Case: Denominator excludes generated code and xtask TEST_0976
status: implemented
verifies: REQ_0994
links incoming: REQ_0994

Inspection of the per-crate summary (target/llvm-cov/summary.txt) after a full run: no rows under target/ (OUT_DIR codegen output) and no xtask rows appear; 49 of the 67 workspace crates carry measurable source. Repeatable as grep -E '(^|/)(target|xtask)/' target/llvm-cov/summary.txt expecting no matches.
Test Case: Missing-tool diagnostic TEST_0977
status: implemented
verifies: REQ_0996
links incoming: REQ_0996

With cargo-llvm-cov absent from PATH, scripts/coverage.sh exits 1 before any build and prints the install hint (cargo install cargo-llvm-cov --locked). Executed 2026-07-03 on a machine without the tool installed, prior to installing it.
Test Case: Contributor documentation present TEST_0978
status: implemented
verifies: REQ_0997
links incoming: REQ_0997

Doc inspection: CONTRIBUTING.md carries a “Test coverage” section under “Building, testing, linting” naming the entrypoint, the install prerequisite, what is measured (all-features, exclusions), the three report locations, and the CI behaviour.
Test Case: CI measures and publishes on code-changing PRs TEST_0979
status: implemented
verifies: REQ_0998, REQ_0999, REQ_1000
links incoming: REQ_0998, REQ_0999, REQ_1000

ci-coverage.yml gates on the shared diff classifier, runs scripts/coverage.sh (the identical local entrypoint), appends the per-crate summary to the GitHub job summary, and uploads lcov.info + the HTML report as the coverage-report artifact with if: !cancelled() so a failing job still publishes. First evidence: actions run 28669729394 on PR #182 (success, artifact uploaded, summary rendered); re-verified by every subsequent run.
Test Case: Coverage floor semantics (gate-ready, unset by default) TEST_0980
status: implemented
verifies: REQ_1001
links incoming: REQ_1001

Against real baseline profile data (87.7% lines): COVERAGE_FAIL_UNDER_LINES unset → exit 0 and summary.txt written; =99 → exit 1 with the lcov and HTML reports still intact (the floor check runs last); =50 → exit 0. Executed 2026-07-03. The workflow ships the variable commented out, so CI today is informational only.