Add option to store local results of Apron-Analysis for comparison

[jerhard]

This PR adds an option to store apron analysis results in a file, such that different analysis runs can be compared using the apronPrivPrecCompare program that is introduced and based on privPrecCompare.

Storing the results of the Apron analysis is activated by setting the string option for "exp.apron-prec-dump" to a non-empty string, which will be used as the file name.

Apron does not offer serialization for its Intervals ("Boxes") and Polyhedra ("Polka").
We thus convert the Apron domain used in the analysis to the octagon domain before serialization. This is done via a conjunction of linear constraints(see here and here, i.e. with to_lincons_array and of_lincons_array).

What is stored / compared

The results for the analysis are stored per program node. Results for different contexts but the same node are joined together. Results of global unknowns are not stored.

What still needs to be done

Storing the results (done in finalize) should only be done in the postsolving stage.
However, when adding such a check, the code is not called. Maybe one needs to change some configuration?

How to test

The comparison program apronPrivPrecCompare apronPrecCompare is built with:

make apronPrecCompare 

The comparison then can be tested on a simple example as follows:

./regtest.sh 36 71 --set ana.apron.domain "octagon" --set exp.apron.prec-dump apron_oct.save
./regtest.sh 36 71 --set ana.apron.domain "interval" --set exp.apron.prec-dump apron_int.save

./apronPrecCompare apron_oct.save apron_int.save

Closes #423

[sim642]

Storing the results (done in finalize) should only be done in the postsolving stage.
However, when adding such a check, the code is not called. Maybe one needs to change some configuration?

finalize is called at the very end of Control anyway and nowhere else (AFAIK), so there doesn't need to be any condition in there. "Postsolving" only refers to the single final evaluation of the entire constraint system and the boolean is actually disabled after that.

[sim642]

Currently the CI fails for all non-apron builds because some util module depends on the Apron stuff. That probably should be separated.

[jerhard]

Currently the CI fails for all non-apron builds because some util module depends on the Apron stuff. That probably should be separated.

The utility module for the apron precision comparison is now moved to a separate module using an optional apron dependency, so that should be fixed.