Publish useful JaCoCo coverage and enforce measured nonregression thresholds #125

Reviewed 2026-09-25 (America/Los_Angeles) against main at bd249f5. Execution order and cross-issue ownership: #169. Batch 04.

This scope replaces the dated implementation prescriptions in the original report and earlier comments; linked historical evidence remains useful but must be rechecked before implementation.

Current scope

The parent prepares JaCoCo data but default verify does not produce/check a report. #162 added Java 8 coverage collection and #152 already removed obsolete benchmark configuration. Old coverage percentages predate new tests and API code.

• Bind report/check goals for relevant library modules independently of the legacy site; skip empty aggregator reports and upload reports through SHA-pinned actions.
• Measure the current baseline, choose justified per-module floors and prove a genuine regression fails. Avoid arbitrary thresholds based on stale line counts.
• Preserve agent configuration, late property evaluation and any other JVM arguments across Surefire/Failsafe; initialize optional properties so skipping coverage still works. The property name itself is not a defect if correctly wired.
• Ensure Java 8 unit-test instrumentation from Add isolated packaged-consumer compatibility CI across Java 8–25 #162 still works and does not inherit an incompatible agent. Do not contaminate isolated packaged-consumer jobs with test agents/libraries; document which integration tests are intentionally instrumented.
• Verify normal and jacoco.skip paths; make report data from distinct JVM runs merge deliberately instead of overwriting it.

#119 owns CI topology, #96 report retirement, #104 plugin pins. PIT/mutation testing remains the separate declined #143 scope and is not silently reintroduced as an acceptance requirement.