Support calling kernel_builder<qreg,...> from CUDA Quantum Kernel

#include "cudaq.h"
#include "cudaq/builder/kernels.h"
#include <iostream>

struct kernel {
  void operator()(auto &&callable) __qpu__ {
    cudaq::qreg q(2);
    callable(q);
  }
};

int main() {

  std::vector<std::complex<double>> state{.70710678, 0., 0., 0.70710678};
  auto [bell, qubits] = cudaq::make_kernel<cudaq::qreg<> &>();
  cudaq::from_state(bell, qubits, state, 2);
  std::cout << bell << "\n";

  kernel test;
  cudaq::sample(test, bell).dump();
}

The issue with this is that a pure-device kernel (bell) created by the kernel_builder does not generate a .thunk function since GenKernelExecution will not add the kernel to the worklist at line 720. So when the callable(q) call happens, kernel_builder::operator()(Args...) is invoked, but fails in the JIT compilation (no thunk, kernelReg, kernelInit, etc), and in invokeCode we'll have no way to call it since there is not thunk.