e2e-nvidia-l40s-x4-llama.yml already uses it. It allows to intelligently find a suitable instance in a number of AZs. Otherwise, a job has a high chance to fail due to insufficient resources.

This issue is to apply the same action to all e2e workflows.