Official build retry failed due to overlapping artifact names with first attempt

Example: https://dev.azure.com/dnceng/internal/_build/results?buildId=2092753&view=logs&j=096efaa6-b2f0-564b-cac3-a8304bac7ae1&t=d709f04b-0994-5a1e-d441-17028e26568f&l=1309

Potential solutions:

• Make sure artifact names have the attempt number in them to ensure retries can succeed. This may involve more complicated YML to access the attempt number in downstream jobs/stages/pipelines.
• Make sure artifact publishes with potentially conflicting names can't happen unless the job/stage is successful. Artifact publish for diagnosis purposes (logs, etc.) can be done in a separate publish step that contains the attempt number. (The attempt number doesn't complicate the build, because downstream jobs/stages don't need to download artifacts that are only uploaded for diagnosis purposes.)
• Upload with attempt number, and re-upload without attempt number later in the build when we know we have had success. This approach might be useful with cross-pipeline dependencies if figuring out the attempt number over the boundary is too complicated.