Make derivative union reduction iterative to avoid stack overflow

The previous commit reduced unions by recursively inserting each disjunct
into the other operand, which recurses with depth proportional to the
union width.  On wide range-product unions (z3test 5721 sub#2) that
overflowed the stack (exit 0xC00000FD), turning a timeout into a crash.

Reformulate mk_union_core to flatten both operands into a disjunct set via
an explicit worklist and reduce it with add_union_elem (a bounded loop
applying subsumption, prefix factoring and same-condition ITE merge
against every existing member).  No width-proportional recursion remains.

5731 stays fixed (0.04s), 5728 stays at ~0.02s, 5721 sub#2 no longer
crashes (cleanly times out as before), 92/92 unit tests pass.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

src/ast/rewriter/seq_derive.h

@@ -163,6 +163,7 @@ namespace seq {
         unsigned union_id(expr* e);              // complement-aware ID for sorting
         bool is_subset(expr* a, expr* b);
         expr_ref mk_union_core(expr* a, expr* b); 
+        void add_union_elem(expr_ref_vector& set, expr* e);
         expr_ref mk_inter(expr* a, expr* b);
         expr_ref mk_inter_core(expr* a, expr* b); 
         expr_ref mk_concat(expr* a, expr* b);