We don't need times for both optimizations enables, but without any optimization
to be able to compare it to the times where only one optimization is enabled.

Makes it easier to use pgfplots

Users can delete them if they want them regenerated

Users can delete them if they want them regenerated

We omitted some fields previously

This substitutes 95%-confidence intervals by the corrected sample standard
deviation. Since we measure different graphs, not the same graph multiple times,
we are less interested in the true mean and error of our measurement than the
variance of the sample. The stddev is the correct measure for this job.

For each set of DFAs for a given size, we now report the of all timing results
instead of all results.

Adds a switch `--dot` to the generator to generate dot files in addition to
valmari and coalgebra files. This makes it easier to compare different graphs
and asses if the generator is working.

Previously we only generated one random DFA per size, which is of course not
statistically relevant. Ten is still not much, but better.