Store external artifacts per pipeline, not per architecture

As we move towards sharing logs upstream, we need to manage upstream artifacts based on the pipeline/test itself rather than by architectures.