Refactor timeseries utilities used in optimization CSVMixin
and PIMixin and move the utilities to a more generic module.

Add a mixin class to enable more seed options,
such as reading a seed from a file
and falling back to a default seed if the solver failed
for the given seed.

Update fill_nan_in_timeseries for the case where the timeseries
has only nans.
In this case, the result should also only be nans.

Call pre- and postprocessing at the correct place in SeedMixin,
so only before and after the actaul optimization.