start/stop verbs: test-fidelity + CLI-suggestion follow-ups (from #2609 review)

Non-blocking follow-ups from the independent review of #2609 (approved + merged).

- **MINOR-1:** CLI unit tests inject a `conduiterr.Wrap` and assert `conduiterr.Get`, but production returns a gRPC `*status.Status`; add one CLI test returning a real `status.New(codes.FailedPrecondition,…).Err()` and assert `exitcode.ExitCode` (higher-fidelity than the mock boundary).
- **MINOR-2:** design test plan promised an end-to-end integration test; PR shipped unit-only (acceptable for Tier-2 wiring over integration-tested RPCs). Add it or mark deferred.
- **MINOR-3:** CLI human error path surfaces the message + exit code but not the structured `suggestion`/`fix` (a pre-existing limitation shared by `inspect` and all client-result commands; MCP delivers full detail). Trim design AC-6's implication that the CLI surfaces the suggestion, or add `FromStatus` rendering to client-result commands generally.

Source: fresh-context review of #2609.

Start with the CLI unit tests that inject conduiterr.Wrap and the production path returning a gRPC status.Status, then inspect the inspect and other client-result commands. Run the CLI tests and check whether an end-to-end integration test exists. Done means the real FailedPrecondition status and exit code are covered, the integration work is added or explicitly deferred, and the suggestion/fix behavior is clarified or implemented consistently.

Written by the indexing model from the issue text.