`paddle.jit.to_static(full_graph=True)` segfaults on `paddle.static.nn.cond` while eager mode works

### bug描述 Describe the Bug `paddle.static.nn.cond` works in eager mode, but wrapping the same logic with `paddle.jit.to_static(full_graph=True)` causes a segmentation fault. ### Minimal reproducible example ```python import paddle a = paddle.ones([1]) def true_fn(): return a + 0.0 def false_fn(): return paddle.zeros([1]) + 0.0 def eager_fn(cond): return paddle.static.nn.cond(cond, true_fn, false_fn) print("eager:", eager_fn(paddle.to_tensor(True))) @paddle.jit.to_static(full_graph=True) def static_fn(cond): return paddle.static.nn.cond(cond, true_fn, false_fn) print("static:", static_fn(paddle.to_tensor(True))) ``` ### Reproduction result Eager mode works normally. to_static(full_graph=True) causes a segmentation fault. ### 其他补充信息 Additional Supplementary Information Paddle version: 3.3