Add optimization to avoid retrieving scale uncertainties when they're not requested

Following this discussion.

For what it's worth, this optimization made negligible difference when I tested it on the ZH analysis; perhaps it will have a greater effect for larger analyses.