feat(TP-195): code-quality typecheck cleanup — npm run typecheck exits 0 on main

[HenryLach]

Fourth of five sequenced packets implementing the code-quality-gates spec. Drives npm run typecheck from 264 errors → 0, satisfying TP-194's CRITICAL pre-condition.

What ships

npm run typecheck exits 0 on this branch.

Type-correctness fixes — large scope

• Runtime source (8 files: types, process-registry, execution, engine, persistence, resume, config-loader, merge, settings-tui, pi-shims): 68 errors fixed via type-drift-only / behavior-neutral changes (most via cascading widening of execLog extras to Record<string, unknown>)
• Tests (~25 files): 196 errors fixed via mock-object schema alignment using new helper factories makeOrchestratorConfig() / makeTaskRunnerConfig()
• Helper module (NEW): extensions/tests/helpers/mock-orchestrator-config.ts deduplicates mock-object construction across 5 large test files

Three bug fixes flagged + endorsed by per-step review

The plan-reviewer (R001 REVISE) flagged that several typecheck errors masked real bugs whose fixes would change runtime behavior. The worker tagged each fix type-drift-only / behavior-neutral / behavior-affecting (escalate) and applied preserve-broken variants by default. The code-reviewer (R003 REVISE) then caught that the preserve-broken casts (undefined as unknown as function stubs, fake structural casts) were themselves shortcuts violating the anti-shortcut policy. Worker applied fix-the-bug variants per R003's explicit recommendation (commit a673a01c); R004 APPROVED the corrected Step 3. Sage independently validated all three fixes as semantically correct.

ID	File	Before	After	Risk
E1	engine.ts:2612	Preflight cleanup Layers 2-5 were stubbed via missing imports (since TP-065). Functions never ran.	Real imports added; Layers 2-5 now execute as originally intended.	Medium — first run after upgrade may clean stale .pi artifacts (intended behavior, not regression).
E2	execution.ts:2924	Read config.failure?.maxWorkerMinutes (camelCase typo, never matched the snake_case schema field). Default 120 always fired.	Read config.failure?.max_worker_minutes. Config now honored.	Medium — operators with explicit max_worker_minutes config see their value take effect (was silently ignored).
E3	resume.ts:2383	Lookup against (batchState as { tasks?: ... }).tasks?.find(...) — non-existent field, always returned undefined. Failure-alert metadata blank for resume-path failures.	Lookup uses laneForTask?.tasks.find(...)?.task — correct source. Alerts now carry segmentIds/activeSegmentId.	Low — IPC schema accepts both populated and blank; populated is strictly more useful.

Plus one cleanup that wasn't behavior-changing:

• extension.ts: dropped 4 dead comparisons to non-existent MonitorState fields (totalDone, totalFailed, currentStep, completedChecks). Comparisons always evaluated true; widget refresh was already only triggering on currentTaskId changes. No behavior change.

Regression tests

3 new behavioral tests verify the fix-the-bug paths (one per bug). Caught by R004's APPROVE review.

Anti-shortcut compliance

• Zero as any casts in the source-side diff (grepped)
• Zero @ts-expect-error / @ts-ignore / // biome-ignore in the source-side diff
• All as unknown as X casts are 2-step legitimate type narrowing (e.g., config-loader.ts property-bag widening, persistence.ts discriminated union helpers)

Sage spot-check confirmed: no shortcut suppression patterns in added hunks.

Pi-shims extension

pi-shims.d.ts extended from ExtensionContext = any to a structural interface exposing typed ui.custom<T>() for the 4 settings-tui.ts call sites. ui is optional so test mocks like { model: null } still satisfy the type. Both @earendil-works/* and @mariozechner/* scope declarations mirror the change.

Validation

Check	Result
npm run typecheck	exit 0 (was: 264 errors)
npm run lint	exit 0 / 280 warnings / 671 infos (warnings and infos non-gating)
npm run format:check	exit 0
npm run test:fast	3627 passing / 1 skipped / 0 failed (+3 from TP-193 baseline of 3624 — the new regression tests)
Sage review	Ship as-is. All fixes semantically correct. Two behavior changes explicitly noted (E1 cleanup activation, E2 config-honored).

Process note (escalation workflow)

The supervisor (me) was unresponsive when the worker's initial escalation fired at the start of Step 1. The two-reviewer loop (R001 plan-review REVISE → re-plan → R002 plan-review APPROVE → Step 3 → R003 code-review REVISE → fix-the-bug → R004 code-review APPROVE) collectively converged on the same outcome the operator would have approved: apply the bug fixes that the typecheck surfaced, with explicit risk documentation. Sage post-merge independently validated each fix. No regression resulted from the unresponsive supervisor, but this is worth noting as evidence that the spec's per-step review pattern provides a real fallback path.

What this does NOT do (per spec scope boundaries)

• ❌ Doesn't enable strict TypeScript mode (Tier-1.5 follow-up)
• ❌ Doesn't change continue-on-error: true in CI (TP-194's job)
• ❌ Doesn't activate the reviewer downgrade rule (TP-194's job)

Closes

(none directly — fourth of five-packet sequence; TP-194 will close the gate-flip work and the inserted-packet rationale documented in the spec)