Compute and output normalized error above tolerance

[stelliom]

This PR adds the computation of a fourth metric for the output of check.py. Instead of only showing the "Error relative to tolerance":

err_rel_tol = err / tol

We also show the "Normalized error above tolerance":

err_above_tol = (err - tol) / tol = err / tol - 1

This let's user see how much above the tolerance the checked variables are, in a normalized fashion, such that errors for different variables can be compared. This is especially useful when err_rel_tol is very close to 1, since the precision of the output only shows 2 decimal digits. With this new output you can see how much above the tolerance the values are even if they are only slightly above (i.e. err_rel_tol ~ 1).

Note that values that do not exceed the tolerance are clipped to 0, this allows for spotting errors quicker.

[stelliom]

@huppd Follow up on our Slack conversation. What do you think?

[huppd]

Thanks for this proposition. I like it :) also the clipping is a good idea.

[huppd]

Suggested change

	"\nError relative to tolerance",
	"\nErrors above tolerance (normalized): max((Differences - Tolerances) / Tolerances, 0)",