Fix example_mwaa_serverless system test race condition with stop_workflow_run (#67578)

The stop_workflow_run task was failing with a ValidationException because
the workflow run completed before the stop command could be issued. This
was particularly problematic on high-latency executors (e.g., ECS) where
task launch overhead (~90s) gave the workflow enough time to finish.

The fix introduces a second workflow definition (stoppable_workflow.yaml)
that sensors for a non-existent S3 key, keeping the workflow running
indefinitely until explicitly stopped.