fix(sparql-anything): stop in-flight processes and clean up when the run is interrupted

[ddeboer]

NativeTaskRunner spawns each task with detached: true and nobody listened for SIGINT/SIGTERM, so interrupting the Node process – Ctrl-C during a long mapping, a cancelled CI job – left the JVMs running and the sparql-anything-* run directory behind.

This takes the converter-level fix from the issue, since the converter is what knows which processes are in flight:

• SparqlAnythingConverter.convert() registers SIGINT/SIGTERM listeners for the duration of a run. On a signal it marks the run as failed (so no further chunk starts), stops the in-flight tasks through the existing stopInFlight() path, removes the run directory, removes its listeners and re-raises the signal with process.kill(process.pid, signal), so the default exit behaviour and exit code are preserved. A second signal while stopping is ignored.
• The listeners are removed in the existing finally, so a completed run leaves none behind. Each run registers its own listener and removes only its own, so parallel convert() calls do not clobber each other.
• NativeTaskRunner’s detached spawning is unchanged: its process-group kill in stop() is what makes stopping work.
• Tests with the package’s fake task runner cover: interruption stops every in-flight task, removes the run directory and the listener before the signal is re-raised; a repeated signal; a run still going after another finished; and no listener left after normal completion. process.kill is stubbed so the test process survives.
• Docs: a paragraph under “Converting several chunks at once”.

Fix #831