[ITSMERNB]

test(cli): guard execute_plan against failed/rebooting steps (#602)

Closes #602.

Thanks to @vladimirrott for catching this gap.

The existing daemon-double tests checked the step verdict, but none exercised the error propagation in the actual execute_plan loop.

This PR extends the existing mock daemon to return a chosen status per action, then runs a real three-step execute_plan for both failure cases:

• failed on the second step: exit code 2.
• needs_reboot on the second step: exit code 5.
• In both cases, the third step is neither previewed nor executed.

Verification on Linux (rust:1-slim):

• cargo fmt --all -- --check — pass.
• cargo test -p sysknife-cli --bins --offline — 321 passed; 0 failed.
• Mutation: replaced exec_result?; with let _ = exec_result;. The new test fails as expected, because the third action executes and the plan returns success:

  test runner::tests::plan_preflight_tests::execute_plan_halts_after_failed_or_rebooting_second_step ... FAILED
  daemon failure/reboot must stop the plan: ()
  test result: FAILED. 0 passed; 1 failed; 320 filtered out
  

• Restored the original line afterward; working tree clean.

This is test-only; no production changes.

[vladimirrott]

Reviewed at 01736334. Approving.

You extended mock_execution_daemon with a per-action status map and left run() as a thin wrapper, so the existing tests that call run() did not change and the new test drives the real execute_plan loop. That was the gap in #602: the earlier tests each wrote their own loop. Asserting the full event sequence, and the step name in the error, pins down both halves of the contract.

What I ran, in a --network=none rust:1-slim container:

filter:    execute_plan_halts_after_failed_or_rebooting_second_step
mutation:  s/^        exec_result?;$/        let _ = exec_result;/
unmutated: 1 executed unit, passing
mutated:   test runner::tests::plan_preflight_tests::execute_plan_halts_after_failed_or_rebooting_second_step ... FAILED (rc=101)

Also checked:

• cargo fmt --all -- --check on the PR tree: rc=0.
• Test count: runner.rs has 101 test attributes on main and 102 here. The diff adds one #[tokio::test] and removes none, so 1,991 is right in the JSON and in all three docs.
• CI: 12 pass, container-smoke skipped.

Nothing blocking and nothing optional. Test-only, so no CHANGELOG line.

If you want the next one, #561 sits beside this code: --dry-run exits 0 for a plan the real run would refuse, because the dry-run path returns before the --max-risk ceiling and the distro check. It is in the pool now. Say so on #561 and I'll hold it for you.