Stop the scan-cancel test racing a fast Release scan

[HeathHowren]

Why

The v2.1.0 merge went red on master: Build and test (Release) failed on test 85, Cancelling a script also cancels the scan it started. Debug passed.

The test was timing-dependent. It submitted a script that scanned, waited and printed, slept 150 ms, cancelled, and then checked the script had not reached its print. An unknown-value scan of the test helper takes a few milliseconds in a Release build, so the script routinely ran to completion before the sleep expired — the cancel arrived with nothing left to cancel, and the print had already happened.

This is not CI-specific. On the same commit, built locally in Release, the old test fails 2 runs in 6. It only ever passed because Debug is slow enough.

What

• The script parks in a loop of its own rather than in scan_wait, so it cannot finish by itself.
• The test waits for scanJob().progress().running to actually become true before cancelling, instead of sleeping and assuming.
• REQUIRE(console.running()) now establishes the premise the old test never checked.
• The final assertion moves from "did the script print" to the scan's own status, which tells a cancelled scan ("Scan cancelled") apart from one that completed. ScanJob::cancel joins its worker, so the scan is over by the time cancel() returns either way — checking running alone could never have failed, and the old test's strongest claim was the one that raced.

No production code changed. This is a test-only fix.

Tests

• Rewritten test run 12× in Release: clean.
• Full suite 5× consecutively in Release (495 test executions): 99/99 every time.
• Debug and Release both build warning-clean under /W4 /WX and pass 99/99.
• Old test re-verified failing 2/6 in Release to confirm the diagnosis rather than assume it.

Needed on master before v2.1.0 can be tagged, since the tag build runs the same suite.