fix: harden TypeScript scanning and plan resolution

[GDownes]

Problem

Four scanner/workflow edge cases surfaced in a TypeScript monorepo:

• unused detection selected the repository-root tsconfig instead of the nearest config owning the scanned project;
• async function body extraction treated object-shaped parameter and return types as executable bodies;
• resolving one queued finding could auto-complete unrelated cluster steps, including steps whose triage refs were summary hashes rather than canonical issue IDs;
• temporarily skipped subjective reassessment prompts still selected the assessment lifecycle phase, then disappeared from display and masked executable triaged findings behind an empty queue.

Fix

• create the temporary unused-check config beside the nearest owning tsconfig;
• scan balanced TypeScript signature delimiters before extracting the executable body;
• evaluate step completion against the whole living plan and fail closed while an unmatched-ref cluster still has members;
• remove skipped subjective items before phase resolution so queued implementation work remains executable.

Validation

• 36 focused queue-snapshot tests passed for the latest fix; the earlier scanner and plan-resolution focused suites also pass.
• Ruff checks and format checks passed for the original six changed files.
• Full suite: 5,662 passed, 152 skipped; three pre-existing Bash source-directive tests fail in this environment.
• Verified all fixes against the WorkOrderGuard services/api scan and living plan; the previously empty queue now exposes the eight triaged review findings and planned coverage work.

[awdemos]

Mixed bundle — two genuinely good fixes, one verified regression, one unsubstantiated claim, some scope creep. Detailed review against main:

Keep:

• step_completion.py fail-closed — real bug, reproduced on main (a step with summary-hash refs auto-completes immediately despite open cluster members); the new test correctly fails on main. One caveat for maintainer sign-off: the all_gone and cluster_issue_ids: continue brush also keeps canonically-referenced steps open until the whole cluster drains — deliberate but broad.
• detector_core._extract_function_body rewrite — broader than fix: harden TypeScript project and async detection #713's async fix (covers arrow functions too, all 4 call sites); index alignment sound. 4 of its 5 new tests fail on main.

Fix or drop:

• unused.py _find_base_tsconfig: verified regression — on a Vite-style root with a {"files": []} solution tsconfig.json plus a real tsconfig.app.json, this picks ./tsconfig.json (tsc compiles nothing → silent empty unused report) where main and fix: harden TypeScript project and async detection #713 pick ./tsconfig.app.json. It also keeps main's silent-empty behavior for tsconfig-less projects that fix: harden TypeScript project and async detection #713 fixes. Recommend taking fix: harden TypeScript project and async detection #713's variant for unused.py (per-directory tsconfig.app.json preference + source fallback) and keeping yours for everything else.
• snapshot.py skipped-subjective filter: the claimed bug does not reproduce on main — the new test passes pre- and post-patch (verified twice, with assessed and unassessed dimension fixtures). A test that passes before and after is not a regression test; make it fail on main or drop the hunk.
• expectTypeOf coverage pattern is fine but unrelated to the stated topic — 4–5 topics in one PR touching risky plan/queue core is scope creep.

Note this textually conflicts with #713 in unused.py — the two PRs fix the same two bugs with different designs (#721 wins on async extraction, #713 wins on tsconfig). They don't stack as-is; consolidating per above is the cleanest path.