#112: Expose canonical inventory, Evidence, and cockpit queues

[justsml]

Split from the Wayfinder mega PR #137.

Refs #112

Scope

• Add the project target-inventory service and API as a canonical public seam.
• Include the durable inventory in project- and thread-scoped cockpit projections.
• Render a read-only Target inventory panel with stable IDs, locators, kinds, authorization state, network profile, safe scope summaries, and relationships.
• Add a project-scoped Artifact detail API with checksum, size, custody links, storage mode, and indexing state.
• Return byte-bounded, redacted text previews while keeping raw and download actions explicitly unavailable.
• Represent binary, missing, and unavailable-storage evidence without attempting unsafe rendering.
• Render active blockers, pending approvals, validation plans, recent findings, and tool failures from the canonical cockpit response.
• Make capped queue totals explicit and preserve full record names through accessible hover text.
• Keep canonical queues read-only; recorded next actions are labels, not stale client-side mutations.
• Suppress the dashboard's historical approval controls when the durable cockpit owns approval state.
• Show compact failure summaries behind an explicit reveal and avoid mislabeling them as full raw tool-run evidence.
• Share the server cockpit DTO with the client so contract drift fails typechecking.
• Treat canonical-record-only projects as dashboard content after reload.

Validation

• pnpm exec vitest run --config vitest.integration.config.ts tests/integration/artifact-preview-api.test.ts tests/integration/artifact-service.test.ts tests/integration/target-inventory-api.test.ts tests/integration/research-cockpit.test.ts (4 files, 16 tests passed before the queue projection)
• Queue follow-up and review repair: cockpit and target-inventory integration files passed (2/2)
• pnpm typecheck
• git diff --check

Review notes

This PR covers the bounded Artifact preview, Target inventory, and distinct canonical cockpit queue slices. It intentionally leaves #112 open: durable blocker storage is #169, full canonical tool-run detail and queue navigation still need their own public routes, and a human UI review remains required.

[justsml]

Council of Dans review: repair/defer. Highest-priority gate: Bound object reads and replace the roughly 18-query cockpit fanout/unbounded authorization-target-blocker reads with paginated aggregate seams and measured query/latency evidence. Feature flags are not a substitute for authority, evidence-integrity, or durability fixes; use typed modes only where they provide a real rollout boundary.

[justsml]

Council triage update: this PR is now draft and parked. It is 391 commits behind main, overlaps newer Artifact/runtime work, and does not yet satisfy #112 as a canonical operator workflow.

P1 blockers:

• object preview reads the full object and does not verify retained bytes; inline writes can silently truncate while reporting full-body integrity;
• no actual raw/download route or audited reveal action;
• canonical queue rows are inert summaries with no detail links, approval actions, diagnostics, or continuation;
• Tool Run failures omit raw input/output, exit state, metadata, ordering, and Artifact links;
• target UI truncates scope and has no canonical detail;
• dashboard still mixes canonical records with chat/prop-derived ownership.

Branch gates are red: stale lock/config, TypeScript failures in the stale environment, Biome/accessibility errors, and no rendered reload/action workflow test. This PR should be rebuilt on current main after Evidence custody settles rather than mechanically merged.