fix: inspect outputs by run kind

[danielgaskins]

Summary

• Detect whether an inspected run is a mounted responsibility or a standalone function before resolving its outputs
• Validate compiled intent, published world models, receipt chains, and versions for mounted responsibilities
• Continue reading declared return bindings for standalone functions
• Report unsupported and contradictory layouts explicitly
• Add focused conformance coverage for both layouts

Use Case / Run Evidence

Fixes #179.

The inspector still expected a retired manifest and treated bindings as the output location for every run. Current mounted responsibility runs instead publish through world-model/ and record commits in receipts/.

This could cause a structurally valid mounted run to be rejected before its outputs were inspected.

Design Boundary

This change updates the standard inspector contract and its focused conformance coverage. It does not change VM semantics, filesystem layout, compilation, or host behavior.

Examples

For a mounted responsibility, the inspector now resolves nodes from compiled-intent.json, reads published outputs from world-model/, and checks them against the receipt ledger.

For a standalone function, the inspector reads declared ### Returns outputs from bindings/ without requiring mounted-only artifacts.

Testing

• corepack pnpm exec vitest run tests/open-prose/evals/inspector.test.ts --maxWorkers=1
• corepack pnpm test -- --maxWorkers=1

The focused suite passed 6 tests. The full suite passed 307 tests across 20 files.

Residual Risk / Follow-ups

The tests verify that the inspector contract follows the documented artifact rules. They do not execute an inspection through a live Prose Complete host.

Issue #180 covers the analogous contract-grader behavior and remains separate.