Was tested on EiV linear and non-EiV linear, but needs further
environment to generate plots.

By this we mean metrics that can only be computed when all seeds are
evaluated so that, in particular, no std can be computed. The only
metric of this kind at this point is the average of the absolute values
of x-dependant biases ('avg-bias'). These metrics will only be computed
for datasets with a dataloader of the type
EIVData.repeated_sampling.repeated_sampling.

This allows now to compute (the average of) a x-dependant bias.

For simulated examples the true coverage is evaluated. For all other
datasets create_tabular will show None.
The coverage was reversed to use the prediction (mean under the
posterior predictive). Note, that the "repeated sampling" assumption is
treated indirectly via averaging over the 10 seeds. For simulated data
this works decently, for real data one often (but not always, cf. wine)
sees a discrepancy.

The script evaluate_tabular.py was renamed to evaluate_metrics. This
script now not only prints the results, but also stores them in JSON
files in a Experiments/results folder (should be created). These files
can be read via a new script create_tabular.py
The JSON files have now all been changed to a less frequent update of
std_y.

Based now on JSON files in results folder. evaluate_tabular.py has been
renamed into evaluate_metrics. JSON files have also been updated. Need
to check whether correct now for all datasets.