Publish whole-workflow language, job and matrix scope #298

Parent

#95. Specification: #295 — Spec: per-run analysis scope and honest finding trends.

What to build

The quality workflow publishes one own-run scope artifact and job summary
combining language presence, selection, detector results, all job/matrix-leg
outcomes and the Layer 2 evidence. A detected but deliberately omitted language
is visibly skipped, while an absent language is not applicable. Failed or
missing discovery is never described as absence.

This step delivers run-level job accounting. Known language-tool entries can
still say execution unknown until the dependent tool-instrumentation step;
a successful job must not stand in for missing tool evidence.

Acceptance criteria

• Presence is observed independently of language selection using the supported project markers. Existing validated target derivation, public detection outputs and refusal behavior remain compatible.
• A fixture containing TypeScript and Python but selecting only TypeScript names Python as present-and-skipped. An absent supported language is not applicable. A deselected unsupported project does not introduce a new detector failure.
• A selected malformed/unsupported project retains the detector failure and diagnostic; dependent jobs are skipped for the failed prerequisite. Independent Layer 2 execution remains represented.
• The collector reconciles expected jobs and matrix legs against received records, including optional test-coverage job state without confusing its percentages with analysis scope.
• Cross-runner transport preserves leg identity and partial records. One successful and one failed matrix leg remain separately visible; a missing fragment is incomplete evidence, not proof of zero examined files or proof the leg never ran.
• Failed-jobs-only reruns have separate attempt identities. Earlier successful records are not silently merged or presented as fresh. Unknown current-attempt execution remains explicit.
• Artifact and summary collection is reachable after ordinary detection, setup and analysis failure. Cancellation metadata is retained when finalization runs; runner loss and forced termination are not promised away.
• Command fixtures exercise discovery, reconciliation and output rendering. Workflow wiring checks assert all dependencies, finalization conditions, artifact transport and non-gating behavior, without claiming live GitHub execution.
• No additional analysis is activated and collection failure cannot change existing gate outcomes or required Adopter configuration.