Fix Unreached Fixpoint in Reluctant Incremental Analysis

[jerhard]

This PR targets issue #949. This PR changes the behavior of reluctant incremental analysis for in case the abstract value of a reluctantly analyzed unknown x changed.

The sets infl x from the previous analysis and the set infl x after the call to solve x by the reluctant analysis are joined before calling destabilize x. Unlike previously, this does not add x to the stable set after the call to destabilize.

[jerhard]

Looking at the old and new infl ret_{foo} (ret_{foo} referring to the unknown at the return node of foo) sets occurring in the reluctant reanalyis of the test case in #949, in turns out that the sets are disjunct: the new set only records the new influence to the newly reachable bar function that calls foo. This is strange, since the infl ret_{foo} set should only grow during a call solve ret_{foo}, since the unknown is in called, and therefore should not be destabilized, and therefore its infl set should not be reset.

It is also not quite clear to me yet why the test case fails when adding the unknown (in this case ret_{foo}) back to the stable set, after destabilizing it.