Support for calling C++ functions within kernels

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

Is calling a "normal" C++ function (say foo) within a kernel a supported operation? I'm not sure what the semantics of that would be... I'm asking because the following:

#include <cudaq.h>

#include <vector>

std::vector<int> foo() { return {0, 1}; }

void kernel() __qpu__ { foo(); }

int main() { cudaq::sample(kernel); }

compiles, but segfaults at runtime. It works fine if the returned value is on the stack (e.g. int, or std::pair).

Interestingly, this pattern (i.e. calling a vector-returning C++ function from within a kernel) is used in our tests in the file unittests/integration/ghz_nisq_tester.cpp, and it works just fine. In fact, the above example runs just fine as a test in our ctest suite...

Steps to reproduce the bug

put the above code in test.cpp. Then run nvq++ test.cpp && ./a.out.

Expected behavior

unclear: either a compilation error or a successful run; not a segfault

Is this a regression? If it is, put the last known working version (or commit) here.

Not a regression

Environment

• CUDA-Q version: main
• Python version: n/a
• C++ compiler: clang-16
• Operating system:

Suggestions

No response

Start with the reproducer in test.cpp and run nvq++ test.cpp && ./a.out to confirm the runtime segfault. Compare it with unittests/integration/ghz_nisq_tester.cpp, where a vector-returning C++ function reportedly works in the ctest suite. Determine whether the supported result should be a compilation error or a successful run, and add coverage for the chosen behavior.