Track uses through first-class modules

[fantazio]

Analyzing code with first-class modules is untested and undocumented.

They are used in Opam 2.5.1 and involved in the following false positives in the UNUSED CONSTRUCTORS/RECORD FIELDS section:

/tmp/proj/opam/src/solver/opamCudfSolverSig.ml:12: criteria_def.crit_default
/tmp/proj/opam/src/solver/opamCudfSolverSig.ml:13: criteria_def.crit_upgrade
/tmp/proj/opam/src/solver/opamCudfSolverSig.ml:14: criteria_def.crit_fixup
/tmp/proj/opam/src/solver/opamCudfSolverSig.ml:15: criteria_def.crit_best_effort_prefix

The fields listed are used in the OpamSolverConfig.with_auto_criteria

[fantazio]

Following discussion in #85, goblint/analyzer uses first-class modules a lot.
In src, the following patterns (typical of first-class modules) appear:

• (module 739 times
• let module 190 times
• (val 57 times

I have not triaged the results of the dead_code_analyzer on that project but, at first glance, there are false positives directly connected to this feature.
The project also uses lots of functors and module inclusions, which will require more robust testing.