Crash with stdexec::task (probably premature coro destruction)

[mika-fischer]

We are seeing this crash relatively reliably on MSVC 17.14.13 (cl 19.44.35215).

I suspect that the task coro handle is destroyed before final_suspend or something along those lines.

I had hoped that the changes in #2038 relating to coro destruction would have fixed this, but unfortunately, it still happens even with those changes.

TEST_CASE("stdexec.task.task_crash_repro") {
    auto task = [] -> stdexec::task<void> { co_return; };
    auto pool = exec::static_thread_pool(1);
    auto scope = stdexec::counting_scope();
    for (int i = 0; i < 1000; ++i) {
        stdexec::spawn(stdexec::starts_on(pool.get_scheduler(), task())
                           | stdexec::upon_error([](auto) noexcept {}),
                       scope.get_token());
    }
    stdexec::sync_wait(scope.join());
}