Pointer flattening: convert typecast through cache

This is to make sure we do not convert the same expression twice, which
would result in fresh variables for accesses into unbounded arrays,
resulting in spurious verification failures.

This was observed with --paths lifo on Pointer_array6 once alignment
assumptions in value-set dereferencing are fixed.

Author: tautschnig

src/solvers/flattening/bv_pointers.cpp

@@ -732,7 +732,7 @@ bvt bv_pointerst::convert_bitvector(const exprt &expr)
     to_typecast_expr(expr).op().type().id() == ID_pointer)
   {
     // pointer to int
-    bvt op0 = convert_pointer_type(to_typecast_expr(expr).op());
+    bvt op0 = convert_bv(to_typecast_expr(expr).op());
 
     // squeeze it in!
     std::size_t width=boolbv_width(expr.type());