[GPU] fix memory conflict for multi iteration in loop. (#28503)

### Details:
 - backport of #28487

### Tickets:
 - 158017

Co-authored-by: Xu, Jia3 <jia3.xu@intel.com>

src/plugins/intel_gpu/src/graph/primitive_inst.cpp

@@ -2391,6 +2391,9 @@ memory::ptr primitive_inst::allocate_output(engine& _engine,
     if (_node.is_in_shape_of_subgraph())
         reusable_across_network = false;
 
+    if (reusable_across_network && _node.get_program().is_body_program() && is_output_buffer && runtime_alloc)
+        reusable_across_network = false;
+
     // For outputs, cpu prim we want to have lockable alloc type
     // Also if the successor of a node is an cpu, then memory needs to be lockable.
     bool is_cpu = _node.get_selected_impl() ? _node.get_selected_impl()->is_cpu() :