`paddle.jit.to_static(full_graph=True)` aborts on `program_guard` with `paddle.equal`

bug描述 Describe the Bug

The following function works in eager mode, but wrapping the same function with paddle.jit.to_static(..., full_graph=True) aborts the Python process.

The failure happens when paddle.static.program_guard is used together with paddle.equal inside the converted function. Even if this mixed static/eager pattern is unsupported, the process should not abort. It would be better to raise a normal Python-level error explaining that this usage is unsupported.

Minimal reproduction

import os

os.environ.setdefault("FLAGS_enable_pir_api", "1")

import paddle


def fn(x):
    main_program = paddle.static.Program()
    startup_program = paddle.static.Program()
    with paddle.static.program_guard(main_program, startup_program):
        y = paddle.equal(x, paddle.zeros_like(x))
    return y, paddle.count_nonzero(y)


x = paddle.to_tensor([[0.0, 1.0], [2.0, 0.0]], dtype="float32")

print("paddle", paddle.__version__, flush=True)
print("eager:", [t.numpy().tolist() for t in fn(x)], flush=True)

static_fn = paddle.jit.to_static(fn, full_graph=True)
out = static_fn(x)
print("static:", [t.numpy().tolist() for t in out], flush=True)

Observed output

paddle 3.3.1
eager: [[[True, False], [False, True]], 2]
EXIT_CODE=-1073740791

W0603 23:21:21.354051 12940 eager_utils.cc:3584] Paddle static graph(PIR) not support input out tensor for now!!!!!
Destroyed a op_result that is still in use by pd_op.cast. The owner op type is: pd_op.equal (at ..\paddle\pir\src\core\op_result_impl.cc:37)

其他补充信息 Additional Supplementary Information

Paddle Version: 3.3.1