fix(runner): fail a parallel run aborted by a malformed annotation

[Chemaclass]

🤔 Background

Related #1335

A malformed @timeout or @retry aborts the file it is in. Sequentially that exits the run, but under --parallel the abort stayed inside the file's worker: alongside one passing file the run printed the error and still reported "All tests passed" with exit 0, so the failure never reached CI.

💡 Changes

• The abort counts as one failed test, which is the question the issue left open. It travels on both channels that cross the fork, because the console summary and the report writers read different ones. Console, json, junit, html and markdown now agree, and the JUnit testcase carries the offending function and the real reason.
• Recorded from the aborting frame, for two reasons. It is the only frame still holding the function and the message, the caller having nothing but an exit status. And under --parallel it is the worker, whose own counter is discarded by design, so each channel contributes exactly once. Recording in the parent instead would land on top of the published payload and re-create --parallel reports count a file-level hook failure twice; the console says one #1301, which a new test guards.
• The wording is built in one place, so the report entry and the stderr line cannot drift apart. Nothing extra is printed to the console: the parent already replays the worker's stderr, and printing again would report one problem twice in two shapes.
• Sequential behaviour, the error text and the exit code are unchanged. Verified on bash 3.2 and bash 5.3.