Record actual language-tool execution inside quality jobs #299

Parent

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

What to build

The quality-run artifact and summary explain which existing language checks
actually ran within each job. They distinguish setup failure, an earlier check
preventing a later one, optional dead-code checks being off or unavailable,
and completed analysis. Job success no longer substitutes for tool execution.

Cover the existing TypeScript lint invocation and knip, .NET build/analyzer
boundary, Python Ruff/mypy/deptry, Rust formatting/Clippy/audit stages, and
Java compiler/analyzer and formatter boundaries. Keep opaque Adopter scripts
opaque rather than inventing evidence about their internal tools.

Acceptance criteria

• Each directly invoked analysis command and known optional tool has its own execution record, outcome/reason and measurement slots. Unsupported numeric counts retain explicit unknown or not-applicable status.
• Record actual npm/pnpm lint-script execution without inferring which internal analyzers ran. Likewise, configured build plugins are not labelled executed solely because their enclosing build succeeded.
• Python fixture failures distinguish environment setup, Ruff execution and mypy prevented by earlier failure. Deptry package-level invocation records respect the existing enumeration boundary.
• Dead-code off, auto with unavailable tool, require with unavailable tool and normal execution are distinguishable without changing their existing verdict behavior.
• Rust formatting failure leaves Clippy and the later audit visibly unexecuted; Clippy failure leaves the later audit unexecuted. Detailed target/feature capture is left to the dependent Rust step.
• Java compilation that ends before the analyzer pass does not claim that pass completed; later formatting prevented by failure is represented separately. Unknown nested execution remains unknown.
• Partial diagnostics survive failure and are referenced from the aggregate artifact and bounded summary. No new unconditional execution is introduced after an existing short-circuit.
• Fixtures run actual workflow command bodies/action handoffs with deterministic success/failure controls; assertions inspect emitted scope and unchanged exit behavior, not just a mirrored state table.
• Workflow/action wiring checks prove records reach the existing aggregate collector on ordinary failure paths. No arbitrary Adopter script rewriting, consumer changes or analysis expansion is included.