Plot normalization with negative central values is wrong

Created by: Zaharid

We normalize errors by doing something like

normalized_error = original_error/central_value

when central_value is negative, this would result in error bars being negative. Previous versions of matplotlib would swallow this and set the error bar to zero, but recent ones complain with

ValueError: 'yerr' must not contain negative values

This could happen when one is computing a data theory comparison with an assymetry, or apparently with arclength plots (see #1657 (closed)).

I believe the fix in the first case is making sure we use the absolute values.

Related #1657 (closed).