[BUG]: is support import numpy in subInterpreter （c++ call python moudule）

[yuetianle]

Required prerequisites

• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker and Discussions to verify that this hasn't already been reported. +1 or comment there if it has.
• Consider asking first in the Gitter chat room or in a Discussion.

What version (or hash if on master) of pybind11 are you using?

3.0.1

Problem description

1. i want use subinterpreter in multithread

CPython version 3.14
pybind11 3.0.1

this is my c++ code
`
PYBIND11_EMBEDDED_MODULE(printer, m, py::multiple_interpreters::per_interpreter_gil()) {
m.def("which", [](const std::string& when) {
std::cout << when << "; Current Interpreter is "
<< py::subinterpreter::current().id()
<< std::endl;
});

}
void worker(int id)
{
try {
// 在 worker 线程内 创建子解释器（必须在该线程销毁）
py::subinterpreter sub = py::subinterpreter::create();

    // 在激活子解释器并持有对应 GIL 的作用域内运行所有 Python 调用
    {
        py::subinterpreter_scoped_activate activate(sub);
        py::gil_scoped_acquire gil; // 必须持有 GIL 才能安全调用 Python

        try {
            // 直接 import 并使用模块 —— 不要把 module 对象带出此作用域
            py::module_::import("printer").attr("which")(
                std::string("worker ") + std::to_string(id) + " : activated subinterpreter"
            );
            py::module::import("numpy");
            printf("========current id:%d", id);

			py::exec(R"(
			import sys
			print("=== sys.path ===")
			for p in sys.path:
				print(p)
			)");
            // 再做一次示例调用（同在激活+GIL作用域内）
            py::module_::import("printer").attr("which")(
                std::string("worker ") + std::to_string(id) + " : second call"
            );

            // 演示抛出 Python 异常并安全捕获（可选）
            // try {
            //     py::module_::import("nonexistent_module"); // 会抛 error_already_set
            // } catch (const py::error_already_set &e) {
            //     std::string msg = e.what(); // safe: 在 GIL + sub interpreter scope
            //     std::cerr << "Caught python error in worker " << id << ": " << msg << '\n';
            // }
        }
        catch (const py::error_already_set &e) {
            // 这里仍然在 sub interpreter + GIL 的作用域内，所以可以安全地调用 e.what()
            std::string msg = e.what();
            std::cerr << "Python exception inside worker " << id << ": " << msg << std::endl;
            // 将 Python 异常信息转成纯 C++ 异常或返回码
        }
        // gil 离开作用域后不能再操作任何 Python 对象
    }

    // sub 在离开作用域后，在同一线程析构 —— 满足 Python 3.12 的 destroy-on-creator-thread 要求
}
catch (const std::exception &ex) {
    std::cerr << "Worker " << id << " std::exception: " << ex.what() << std::endl;
}

}
int test_main_thread()
{
py::scoped_interpreter guard{};
// 在主线程里可以做一些主解释器的调用（如果需要）
{
py::gil_scoped_acquire gil;
py::module_::import("printer").attr("which")("main: before starting worker threads");
}
py::gil_scoped_release release;

std::thread t1(worker, 1);
std::thread t2(worker, 2);


t1.join();
t2.join();

}`

printer can excute but numpy give me such tips
Python exception inside worker 1: ModuleNotFoundError: No module named 'numpy'
Python exception inside worker 2: ModuleNotFoundError: No module named 'numpy'

when i use these code

py::exec(R"(
import sys
sys.path.append(r"C:\Users\shida\AppData\Local\Programs\Python\Python312\Lib")
sys.path.append(r"C:\Users\shida\AppData\Local\Programs\Python\Python312\Lib\site-packages")
)");

append the numpy lib in sys.path give me can't use in source directory

Reproducible example code

Is this a regression? Put the last known working version here if it is.

Not a regression