Conditional sampling detection not working for C++ lambda kernels

cudaq::kernelHasConditionalFeedback doesn't seem to work properly for kernels written as lambdas and potentially C-style functions as well.

This may lead to incorrect simulation mode being selected.

Reproducer:

#include <cudaq.h>
#include <iostream>

struct kernel_struct {
  void operator()() __qpu__ {
    cudaq::qubit q;
    h(q);
    if (mz(q))
      x(q);
    mz(q);
  }
};

int main() {
  auto kernel_lambda = []() __qpu__ {
    cudaq::qubit q;
    h(q);
    if (mz(q))
      x(q);
    mz(q);
  };

  std::cout << "Counts from kernel written as a struct:\n";
  cudaq::sample(kernel_struct{}).dump();

  std::cout << "Counts from kernel written as a lambda:\n";
  cudaq::sample(kernel_lambda).dump();
}

The results for the two kernels are different. The one written as a struct produced a correct result:

Counts from kernel written as a struct:
{ 0:1000 }
Counts from kernel written as a lambda:
{ 0:501 1:499 }

I suspected that it may have failed to look up the quake code from the registry.

Start with the supplied C++ reproducer and inspect cudaq::kernelHasConditionalFeedback, focusing on how lambda kernels are looked up in the quake code registry. Compare the detection path for the lambda and kernel_struct cases. Done means both kernels select the same correct simulation behavior and produce matching results.