fix(lab): a failed run stops the Lab's sweep, a run cancelled while starting is cancelled on the server, and diff rows keep their order

[bmscomp]

Item 1.8 of the CLI performance and stability plan (findings F24 and F40's Lab half). Independent of the other Wave 1 PRs.

Why

Three faults in kates lab, on main (6f4cf61):

• A failed run counts as a measurement (cli/tui/lab.go:1115-1117). pollTestOnce sends FAILED and ERROR as an ordinary done message, so the run becomes a 0 rec/s iteration. A sweep, warmup or median series then launches its next step against the same failure.
• A run cancelled while it starts keeps running. Pressing x while CreateTest is in flight sets the Lab not running. The labTestStartedMsg that follows is then dropped (:396-398), and the run the API started goes on with nothing watching or cancelling it.
• The diff rows reorder on every frame (:994-1004). The Parameter Changes rows range over a map.

Changes

cli/tui/lab.go:

• A FAILED or ERROR status returns a done message with an error (test <id> ended FAILED). The Lab's existing error path takes it: no iteration is recorded, the sweep, warmup and median series stop, and the status line offers r to retry.
• A labTestStartedMsg that arrives when the Lab is no longer running cancels its run on the server. The command that does it is the new cancelRun, which the x key now uses too, so both send the same 10 s-bounded POST /api/tests/{id}/cancel.
• The diff's parameter rows are in key order (slices.Sorted(maps.Keys(…))).

Tests

Added:

• TestLabFailedRunStopsTheSweep, for FAILED and ERROR in parallel: pollTestOnce against a stub returning that status gives an error, and in a running sweep it records no iteration, stops the sweep, and names the status. It takes 2 s, pollTestOnce's own wait.
• TestLabDiffParameterRowsKeepTheirOrder: 20 renders of a five-key diff are identical, and the keys are in sorted order.

Changed:

• TestLabTestStartedMsg's "dropped when the run was cancelled during creation" case expected no command, which is the bug. It is now "cancelled on the server…", and it sees POST /api/tests/run-42/cancel reach a stub.

On main's lab.go all three fail: no cancel command, a nil error for FAILED and ERROR, and a diff that changed between frames.

Run locally, passing:

• cd cli && go test ./..., go test -race ./tui, go vet ./....
• golangci-lint run --new-from-rev=origin/main ./...: 0 issues.
• scripts/check-cli-style.sh and scripts/check-cli-compat.sh: OK.
• The backend suite, which cli/** triggers (Build & Test): this PR touches nothing under kates/. ./mvnw test on main's kates/ tree ran 1,254 tests with 0 failures and 0 errors, once for all of Wave 1.

Skipped locally: govulncheck, as go.mod is unchanged.

CI only: ./mvnw verify with the *IT tests, the Kind and Docker jobs, CodeQL and Trivy.

Deliberately not done

• F40's other half, a topic filter with no match showing every topic (cli/tui/topics.go:90-108), is item 6.2.
• A started message for an earlier create that arrives after the Lab has started a newer run is still taken as the current run. That needs x, then a fresh start, both inside one CreateTest round trip. The plan does not list it, and it is left out.