Follow-up crash fixes to Apron and invariants

[sim642]

This PR consists of a small set of fixes I made locally when I merged together #759, #762 and #742, and tried running the combination on pthread bench. Now that all three are merged, I can submit these.

• The change of get_ikind is related to that now being used in Apron on mistyped alloc variables. Otherwise those void variables got the default IInt kind, which caused an incompatible ikinds crash in the meet of EvalInt query results.
• The change of mkCast is similar but applies to the cast being added in base int invariants.

[sim642]

This default IInt ikind caused some mismatches in query result meets. Here I've replaced it with an exception and all our tests still pass. Although I don't know why the IInt default existed, as opposed to an exception. get_ikind (and get_ikind_exp) are somewhat extensively used, which made me worry about changing this.

[jerhard]

It seems reasonable to throw an exception when get_ikind is applied to a type that is not a int, enum, or pointer. In case we will encounter some uncaught exceptions caused by this in the future, there probably needs to be some handling specific to the call-site of get_ikind, anyway.

[sim642]

Exactly! Having the exception immediately here makes it much easier to track down from the backtrace as opposed to the ikind flowing through bunch of other code before crashing at an unrelated place.